Machine learning for activity recognition: hip versus wrist data

Recognition of sports and daily activities through deep learning and convolutional block attention

Portable devices like accelerometers and physiological trackers capture movement and biometric data relevant to sports. This study uses data from wearable sensors to investigate deep learning techniques for recognizing human behaviors associated with sports and fitness. The proposed CNN-BiGRU-CBAM model, a unique hybrid architecture, combines convolutional neural networks (CNNs), bidirectional gated recurrent unit networks (BiGRUs), and convolutional block attention modules (CBAMs) for accurate activity recognition. CNN layers extract spatial patterns, BiGRU captures temporal context, and CBAM focuses on informative BiGRU features, enabling precise activity pattern identification. The novelty lies in seamlessly integrating these components to learn spatial and temporal relationships, prioritizing significant features for activity detection. The model and baseline deep learning models were trained on the UCI-DSA dataset, evaluating with 5-fold cross-validation, including multi-class classification accuracy, precision, recall, and F1-score. The CNN-BiGRU-CBAM model outperformed baseline models like CNN, LSTM, BiLSTM, GRU, and BiGRU, achieving state-of-the-art results with 99.10% accuracy and F1-score across all activity classes. This breakthrough enables accurate identification of sports and everyday activities using simplified wearables and advanced deep learning techniques, facilitating athlete monitoring, technique feedback, and injury risk detection. The proposed model's design and thorough evaluation significantly advance human activity recognition for sports and fitness.

Examining the Association Between Objective Physical Activity and Momentary Pain: A Systematic Review of Studies Using Ambulatory Assessment

Chronic pain (CP) is a debilitating and increasingly common health condition that adversely impacts function, including physical activity (PA). Research using ambulatory assessment (AA) methods (eg, ecological momentary assessment, actigraphy) offers promise for elucidating the relationship between momentary pain and objective PA in CP populations. This study aimed to systematically review articles assessing the association between momentary pain and PA in adults with CP as measured using AA and to make recommendations for the measurement and study of this relationship. Five databases were systematically searched, and 13 unique records (N = 768) met the inclusion criteria. CP conditions included mixed/nonspecific CP (k = 3), low back pain (k = 2), fibromyalgia (k = 1), unspecified arthritis (k = 1), and hip/knee osteoarthritis (k = 6). The average age of participants across studies was 55.29 years, and the majority identified as women (60.68%) and White (83.16%). All studies measured objective PA via actigraphy, and momentary pain with either a diary/log or ratings on an actigraph. Studies varied in the quantification of PA (ie, activity counts, step count, moderate-vigorous PA), statistical method (ie, correlation, regression, multilevel modeling), and inclusion of moderators (eg, pain acceptance). Studies reported mixed results for the pain-PA relationship. This heterogeneity suggests that no summarizing conclusions can be drawn about the pain-PA relationship without further investigation into its complex nuances. More within-person and exploratory examinations that maximize the richness of AA data are needed. A greater understanding of this relationship can inform psychotherapeutic and behavioral recommendations to improve CP outcomes. PERSPECTIVE: This article presents a systematic review of the literature on the association between momentary pain and PA in adults with CP as measured using AA methods. A better understanding of this nuanced relationship could help elucidate areas for timely intervention and may inform clinical recommendations to improve CP outcomes. PROSPERO REGISTRATION NUMBER: CRD42023389913.

Concurrent Validity of Four Activity Monitors in Older Adults

Sedentary behaviour (SB) and physical activity (PA) have been shown to be independent modulators of healthy ageing. We thus investigated the impact of activity monitor placement on the accuracy of detecting SB and PA in older adults, as well as a novel random forest algorithm trained on data from older persons. Four monitor types (ActiGraph wGT3X-BT, ActivPAL3c VT, GENEActiv Original, and DynaPort MM+) were simultaneously worn on five anatomical sites during ten different activities by a sample of twenty older adults (70.0 (12.0) years; 10 women). The results indicated that collecting metabolic equivalent (MET) data for 60 s provided the most representative results, minimising variability. In addition, thigh-worn monitors, including ActivPAL, Random Forest, and Sedentary Sphere-Thigh, exhibited superior performance in classifying SB, with balanced accuracies ≥ 94.2%. Other monitors, such as ActiGraph, DynaPort MM+, and GENEActiv Sedentary Sphere-Wrist, demonstrated lower performance. ActivPAL and GENEActiv Random Forest outperformed other monitors in participant-specific balanced accuracies for SB classification. Only thigh-worn monitors achieved acceptable overall balanced accuracies (≥80.0%) for SB, standing, and medium-to-vigorous PA classifications. In conclusion, it is advisable to position accelerometers on the thigh, collect MET data for ≥60 s, and ideally utilise population-specific trained algorithms.

Can Gait Characteristics Be Represented by Physical Activity Measured with Wrist-Worn Accelerometers?

Wearable accelerometers allow for continuous monitoring of function and behaviors in the participant's naturalistic environment. Devices are typically worn in different body locations depending on the concept of interest and endpoint under investigation. The lumbar and wrist are commonly used locations: devices placed at the lumbar region enable the derivation of spatio-temporal characteristics of gait, while wrist-worn devices provide measurements of overall physical activity (PA). Deploying multiple devices in clinical trial settings leads to higher patient burden negatively impacting compliance and data quality and increases the operational complexity of the trial. In this work, we evaluated the joint information shared by features derived from the lumbar and wrist devices to assess whether gait characteristics can be adequately represented by PA measured with wrist-worn devices. Data collected at the Pfizer Innovation Research (PfIRe) Lab were used as a real data example, which had around 7 days of continuous at-home data from wrist- and lumbar-worn devices (GENEActiv) obtained from a group of healthy participants. The relationship between wrist- and lumbar-derived features was estimated using multiple statistical methods, including penalized regression, principal component regression, partial least square regression, and joint and individual variation explained (JIVE). By considering multilevel models, both between- and within-subject effects were taken into account. This work demonstrated that selected gait features, which are typically measured with lumbar-worn devices, can be represented by PA features measured with wrist-worn devices, which provides preliminary evidence to reduce the number of devices needed in clinical trials and to increase patients' comfort. Moreover, the statistical methods used in this work provided an analytic framework to compare repeated measures collected from multiple data modalities.

The effects of active classroom breaks on moderate to vigorous physical activity, behaviour and performance in a Northern NSW primary school: A quasi-experimental study

ISSUE ADDRESSED

Approximately 77% of NSW children aged 5 to 15 years do not meet physical activity guidelines and many spend a considerable amount of time sitting. Active breaks at primary school are feasible, may increase daily moderate to vigorous physical activity (MVPA) and decrease off-task behaviour without adversely affecting cognitive function and learning.

METHODS

In this quasi-experimental study, 101 primary school children in six intervention classrooms participated in three 10-minute active breaks per day for six-weeks during class time, while five control classrooms were run as usual (n = 89). Physical activity levels were measured using wrist-worn Actigraph wGT3X-BT accelerometers and analysed using a random forest model. Students' off-task behaviour, wellbeing, cognitive function and maths performance were also measured. School staff completed a brief feedback survey.

RESULTS

Children in the intervention group engaged in 15.4 and 10.9 minutes more MVPA per day at 3 and 6 weeks respectively (P < .001). Participation significantly increased the proportion of children who met the Australian 24-Hour Movement Guidelines (P < .001). At pre, middle and end of intervention, 44.4%, 60.8% and 55.1% of intervention children and 46.5%, 45.9% and 45.8% of controls met the guidelines. Significantly fewer students engaged in off-task behaviour in the intervention classes at mid and final weeks of intervention (-1.4 students, P = .003). No significant intervention effects were found for wellbeing, cognitive and maths performance.

CONCLUSIONS

Active classroom breaks are an effective way to increase physical activity among primary school children while reducing off-task classroom behaviour. SO WHAT?: Primary school students' health would benefit from active breaks with no detrimental effects on wellbeing, maths and cognitive performance.

Hang-Time HAR: A Benchmark Dataset for Basketball Activity Recognition Using Wrist-Worn Inertial Sensors

We present a benchmark dataset for evaluating physical human activity recognition methods from wrist-worn sensors, for the specific setting of basketball training, drills, and games. Basketball activities lend themselves well for measurement by wrist-worn inertial sensors, and systems that are able to detect such sport-relevant activities could be used in applications of game analysis, guided training, and personal physical activity tracking. The dataset was recorded from two teams in separate countries (USA and Germany) with a total of 24 players who wore an inertial sensor on their wrist, during both a repetitive basketball training session and a game. Particular features of this dataset include an inherent variance through cultural differences in game rules and styles as the data was recorded in two countries, as well as different sport skill levels since the participants were heterogeneous in terms of prior basketball experience. We illustrate the dataset's features in several time-series analyses and report on a baseline classification performance study with two state-of-the-art deep learning architectures.

Wearable accelerometers for measuring and monitoring the motor behaviour of infants with brain damage during CareToy-Revised training

BACKGROUND

Nowadays, wearable sensors are widely used to quantify physical and motor activity during daily life, and they also represent innovative solutions for healthcare. In the clinical framework, the assessment of motor behaviour is entrusted to clinical scales, but they are dependent on operator experience. Thanks to their intrinsic objectivity, sensor data are extremely useful to provide support to clinicians. Moreover, wearable sensors are user-friendly and compliant to be used in an ecological environment (i.e., at home). This paper aims to propose an innovative approach useful to predict clinical assessment scores of infants' motor activity.

MATERIALS AND METHODS

Starting from data acquired by accelerometers placed on infants' wrists and trunk during playtime, we exploit the method of functional data analysis to implement new models combining quantitative data and clinical scales. In particular, acceleration data, transformed into activity indexes and combined with baseline clinical data, represent the input dataset for functional linear models.

CONCLUSIONS

Despite the small number of data samples available, results show correlation between clinical outcome and quantitative predictors, indicating that functional linear models could be able to predict the clinical evaluation. Future works will focus on a more refined and robust application of the proposed method, based on the acquisition of more data for validating the presented models.

TRIAL REGISTRATION NUMBER

ClincalTrials.gov; NCT03211533. Registered: July, 7th 2017. ClincalTrials.gov; NCT03234959. Registered: August, 1st 2017.

Wearable Motion Capture Devices for the Prevention of Work-Related Musculoskeletal Disorders in Ergonomics-An Overview of Current Applications, Challenges, and Future Opportunities

Work-related musculoskeletal disorders (WMSDs) are a major contributor to disability worldwide and substantial societal costs. The use of wearable motion capture instruments has a role in preventing WMSDs by contributing to improvements in exposure and risk assessment and potentially improved effectiveness in work technique training. Given the versatile potential for wearables, this article aims to provide an overview of their application related to the prevention of WMSDs of the trunk and upper limbs and discusses challenges for the technology to support prevention measures and future opportunities, including future research needs. The relevant literature was identified from a screening of recent systematic literature reviews and overviews, and more recent studies were identified by a literature search using the Web of Science platform. Wearable technology enables continuous measurements of multiple body segments of superior accuracy and precision compared to observational tools. The technology also enables real-time visualization of exposures, automatic analyses, and real-time feedback to the user. While miniaturization and improved usability and wearability can expand the use also to more occupational settings and increase use among occupational safety and health practitioners, several fundamental challenges remain to be resolved. The future opportunities of increased usage of wearable motion capture devices for the prevention of work-related musculoskeletal disorders may require more international collaborations for creating common standards for measurements, analyses, and exposure metrics, which can be related to epidemiologically based risk categories for work-related musculoskeletal disorders.

Is Wrist Accelerometry Suitable for Threshold Scoring? A Comparison of Hip-Worn and Wrist-Worn ActiGraph Data in Low-Active Older Adults With Obesity

BACKGROUND

Hip- and wrist-worn ActiGraph accelerometers are widely used in research on physical activity as they offer an objective assessment of movement intensity across the day. Herein we characterize and contrast key structured physical activities and common activities of daily living via accelerometry data collected at the hip and wrist from a sample of community-dwelling older adults.

METHODS

Low-active, older adults with obesity (age 60+ years) were fit with an ActiGraph GT3X+ accelerometer on their nondominant wrist and hip before completing a series of tasks in a randomized order, including sitting/standing, sweeping, folding laundry, stair climbing, ambulation at different intensities, and cycling at different intensities. Participants returned a week later and completed the tasks once again. Vector magnitude counts/second were time-matched during each task and then summarized into counts/minute (CPM).

RESULTS

Monitors at both wear locations similarly characterized standing, sitting, and ambulatory tasks. A key finding was that light home chores (sweeping, folding laundry) produced higher and more variable CPM values than fast walking via wrist ActiGraph. Regression analyses revealed wrist CPM values were poor predictors of hip CPM values, with devices aligning best during fast walking (R2 = 0.25) and stair climbing (R2 = 0.35).

CONCLUSIONS

As older adults spend a considerable portion of their day in nonexercise activities of daily living, researchers should be cautious in the use of simply acceleration thresholds for scoring wrist-worn accelerometer data. Methods for better classifying wrist-worn activity monitor data in older adults are needed.

Conceptual Structure and Current Trends in Artificial Intelligence, Machine Learning, and Deep Learning Research in Sports: A Bibliometric Review

Artificial intelligence and its subcategories of machine learning and deep learning are gaining increasing importance and attention in the context of sports research. This has also meant that the number of corresponding publications has become complex and unmanageably large in human terms. In the current state of the research field, there is a lack of bibliometric analysis, which would prove useful for obtaining insights into the large amounts of available literature. Therefore, the present work aims to identify important research issues, elucidate the conceptual structure of the research field, and unpack the evolutionary trends and the direction of hot topics regarding key themes in the research field of artificial intelligence in sports. Using the Scopus database, 1215 documents (reviews and articles) were selected. Bibliometric analysis was performed using VOSviewer and bibliometrix R package. The main findings are as follows: (a) the literature and research interest concerning AI and its subcategories is growing exponentially; (b) the top 20 most cited works comprise 32.52% of the total citations; (c) the top 10 journals are responsible for 28.64% of all published documents; (d) strong collaborative relationships are present, along with small, isolated collaboration networks of individual institutions; (e) the three most productive countries are China, the USA, and Germany; (f) different research themes can be characterized using author keywords with current trend topics, e.g., in the fields of biomechanics, injury prevention or prediction, new algorithms, and learning approaches. AI research activities in the fields of sports pedagogy, sports sociology, and sports economics seem to have played a subordinate role thus far. Overall, the findings of this study expand knowledge on the research situation as well as the development of research topics regarding the use of artificial intelligence in sports, and may guide researchers to identify currently relevant topics and gaps in the research.

Exercise intensity of active video gaming in cerebral palsy: hip- versus wrist-worn accelerometer data

OBJECTIVE

The aim of this study was to compare exercise intensity of active video games (AVGs) between hip- and wrist-worn accelerometer data in cerebral palsy (CP).

METHODS

Twenty children and adolescents (9.35 ± 3.71 years) with CP performed two exercise sessions, completing a standardized series of AVGs. Exercise intensity was collected, while one accelerometer was fitted to wrist and hip in separate, counterbalanced sessions.

RESULTS

Accelerometer counts per minute and cut-points determined were significantly different between the wrist- and hip-worn outputs (p < .001). Metabolic equivalents (METs) of performing AVGs exceeded the three METs moderate intensity threshold in wrist-worn (3.12 ± 0.86) accelerometer and hip-worn data tend to underestimate intensity (1.16 ± 0.08).

CONCLUSIONS

Previous studies showed METs required to perform AVGs were related to moderate intensity (3-6 METs) in CP with mild deficits. Wrist-worn accelerometer, exceeding 3 METs, seem to have higher accuracy in measuring exercise intensity of AVGs than hip-worn.

Physical activity, memory function, and hippocampal volume in adults with Down syndrome

Higher engagement in moderate-intensity physical activity (PA) is related to better cognitive functioning in neurotypical adults; however, little is known about the effect of PA on cognitive aging in adults with Down syndrome (DS). Individuals with DS have three copies of chromosome 21, which includes the gene involved in the production of the amyloid precursor protein, resulting in an increased risk for an earlier onset of Alzheimer’s disease (AD). The goal of this study was to understand the relationship between engagement in moderate PA, memory, and hippocampal volume in adults with DS. Adults with DS participated in an ancillary Lifestyle study linked to the Alzheimer’s Biomarkers Consortium for DS (ABC- DS; N = 71). A within-sample z-score memory composite was created from performance on the Cued Recall Test (CRT) and the Rivermead Picture Recognition Test. Participants wore a wrist-worn accelerometer (GT9X) to measure PA. Variables of interest included the average percentage of time spent in moderate PA and average daily steps. Structural MRI data were acquired within 18 months of actigraphy/cognitive data collection for a subset of participants (n = 54). Hippocampal volume was extracted using Freesurfer v5.3. Associations between moderate PA engagement, memory, and hippocampal volume were evaluated with hierarchical linear regressions controlling for relevant covariates [age, body mass index, intellectual disability level, sex, and intracranial volume]. Participants were 37.77 years old (SD = 8.21) and were 55.6% female. They spent 11.1% of their time engaged in moderate PA (SD = 7.5%) and took an average of 12,096.51 daily steps (SD = 4,315.66). After controlling for relevant covariates, higher memory composite score was associated with greater moderate PA engagement (β = 0.232, p = 0.027) and more daily steps (β = 0.209, p = 0.037). In a subset of participants, after controlling for relevant covariates, PA variables were not significantly associated with the hippocampal volume (all p-values ≥ 0.42). Greater hippocampal volume was associated with higher memory composite score after controlling for relevant covariates (β = 0.316, p = 0.017). More PA engagement was related to better memory function in adults with DS. While greater hippocampal volume was related to better memory performance, it was not associated with PA. Greater PA engagement may be a promising lifestyle behavior to preserve memory in adults with DS.

Concurrent validity of the combined HRV/ACC sensor and physical activity diary when monitoring physical activity in university students during free-living days

The purpose of this research was to determine if the scientific research device combined heart rate variability combined with an acceleration sensor (Firstbeat Bodyguard 2, BG2) was valid and reliable for time spent in different intensity zones in free-living. A total of 55 healthy participants performed 48-h physical activity (PA) monitoring with BG2, ActiGraph GT3X+ (GT3X+), and completed Bouchard Physical Activity Diary (Bouchard) every night. In the available studies, GT3X+ is considered the gold standard scientific research device for PA monitor. We compared BG2 and Bouchard with GT3X+ by difference, correlation, and agreement of PA and energy expenditure (EE) in free-living. The results showed that BG2 estimated PA more accurately than Bouchard, with a modest correlation (r > 0.49), strong agreement (τ > 0.29), and they had the lowest limits of agreement when estimating moderate to vigorous physical activity (MVPA). The EE estimated by Bouchard was the highest among the three methods, and the correlation and agreement between the three methods were high. Our findings showed that the BG2 is valid and reliable for estimating time spent in different intensity zones in free-living, especially in MVPA.

A novel smartphone-based activity recognition modeling method for tracked equipment in forest operations

Activity recognition modelling using smartphone Inertial Measurement Units (IMUs) is an underutilized resource defining and assessing work efficiency for a wide range of natural resource management tasks. This study focused on the initial development and validation of a smartphone-based activity recognition system for excavator-based mastication equipment working in Ponderosa pine (Pinus ponderosa) plantations in North Idaho, USA. During mastication treatments, sensor data from smartphone gyroscopes, accelerometers, and sound pressure meters (decibel meters) were collected at three sampling frequencies (10, 20, and 50 hertz (Hz)). These data were then separated into 9 time domain features using 4 sliding window widths (1, 5, 7.5 and 10 seconds) and two levels of window overlap (50% and 90%). Random forest machine learning algorithms were trained and evaluated for 40 combinations of model parameters to determine the best combination of parameters. 5 work elements (masticate, clear, move, travel, and delay) were classified with the performance metrics for individual elements of the best model (50 Hz, 10 second window, 90% window overlap) falling within the following ranges: area under the curve (AUC) (95.0% - 99.9%); sensitivity (74.9% - 95.6%); specificity (90.8% - 99.9%); precision (81.1% - 98.3%); F1-score (81.9% - 96.9%); balanced accuracy (87.4% - 97.7%). Smartphone sensors effectively characterized individual work elements of mechanical fuel treatments. This study is the first example of developing a smartphone-based activity recognition model for ground-based forest equipment. The continued development and dissemination of smartphone-based activity recognition models may assist land managers and operators with ubiquitous, manufacturer-independent systems for continuous and automated time study and production analysis for mechanized forest operations.

Bronchiectasis - Exercise as Therapy (BREATH): rationale and study protocol for a multi-center randomized controlled trial

Background

Globally, bronchiectasis (BE) unrelated to cystic fibrosis (CF) is recognized as a major cause of respiratory morbidity, mortality, and healthcare utilization. Children with BE regularly experience exacerbations of their condition resulting in frequent hospitalizations and decreased health-related quality of life (HR-QoL). Guidelines for the treatment and management of BE call for regular exercise as a means of improving aerobic fitness and HR-QoL. Moreover, research in adults with BE has shown that exercise can reduce the frequency of exacerbations, a potent predictor of future lung function decline and respiratory morbidity. Yet, to date, the health benefits resulting from therapeutic exercise have not been investigated in children with BE. The BREATH, Bronchiectasis - Exercise as Therapy, trial will test the efficacy of a novel 8-week, play-based therapeutic exercise program to reduce the frequency of acute exacerbations over 12 months in children with BE (aged ≥ 4 and < 13 years). Secondary aims are to determine the cost-effectiveness of the intervention and assess the program’s impact on aerobic fitness, fundamental movement skill (FMS) proficiency, habitual physical activity, HR-QoL, and lung function.

Methods

This multi-center, observer-blinded, parallel-group (1:1 allocation), randomized controlled trial (RCT) will be conducted at three sites. One hundred and seventy-four children ≥ 4 and < 13 years of age with BE will be randomized to a developmentally appropriate, play-based therapeutic exercise program (eight, 60-min weekly sessions, supplemented by a home-based program) or usual care. After completing the baseline assessments, the number of exacerbations and secondary outcomes will be assessed immediately post-intervention, after 6 months of follow-up, and after 12 months of follow-up. Monthly, parental contact and medical review will document acute respiratory exacerbations and parameters for cost-effectiveness outcomes.

Discussion

The BREATH trial is the first fully powered RCT to test the effects of a therapeutic exercise on exacerbation frequency, fitness, movement competence, and HR-QoL in children with bronchiectasis. By implementing a developmentally appropriate, play-based exercise program tailored to the individual needs of children with bronchiectasis, the results have the potential for a major paradigm shift in the way in which therapeutic exercise is prescribed and implemented in children with chronic respiratory conditions. The exercise program can be readily translated. It does not require expensive equipment and can be delivered in a variety of settings, including the participant’s home. The program has strong potential for translation to other pediatric patient groups with similar needs for exercise therapy, including those with obesity, childhood cancers, and neurological conditions such as cerebral palsy.

Trial registration

Australian and New Zealand Clinical Trials Register (ANZCTR) ACTRN12619001008112

Robust Human Activity Recognition by Integrating Image and Accelerometer Sensor Data Using Deep Fusion Network

Studies on deep-learning-based behavioral pattern recognition have recently received considerable attention. However, if there are insufficient data and the activity to be identified is changed, a robust deep learning model cannot be created. This work contributes a generalized deep learning model that is robust to noise not dependent on input signals by extracting features through a deep learning model for each heterogeneous input signal that can maintain performance while minimizing preprocessing of the input signal. We propose a hybrid deep learning model that takes heterogeneous sensor data, an acceleration sensor, and an image as inputs. For accelerometer data, we use a convolutional neural network (CNN) and convolutional block attention module models (CBAM), and apply bidirectional long short-term memory and a residual neural network. The overall accuracy was 94.8% with a skeleton image and accelerometer data, and 93.1% with a skeleton image, coordinates, and accelerometer data after evaluating nine behaviors using the Berkeley Multimodal Human Action Database (MHAD). Furthermore, the accuracy of the investigation was revealed to be 93.4% with inverted images and 93.2% with white noise added to the accelerometer data. Testing with data that included inversion and noise data indicated that the suggested model was robust, with a performance deterioration of approximately 1%.

The Dilemma of Analyzing Physical Activity and Sedentary Behavior with Wrist Accelerometer Data: Challenges and Opportunities

Physical behaviors (e.g., physical activity and sedentary behavior) have been the focus among many researchers in the biomedical and behavioral science fields. The recent shift from hip- to wrist-worn accelerometers in these fields has signaled the need to develop novel approaches to process raw acceleration data of physical activity and sedentary behavior. However, there is currently no consensus regarding the best practices for analyzing wrist-worn accelerometer data to accurately predict individuals' energy expenditure and the times spent in different intensities of free-living physical activity and sedentary behavior. To this end, accurately analyzing and interpreting wrist-worn accelerometer data has become a major challenge facing many clinicians and researchers. In response, this paper attempts to review different methodologies for analyzing wrist-worn accelerometer data and offer cutting edge, yet appropriate analysis plans for wrist-worn accelerometer data in the assessment of physical behavior. In this paper, we first discuss the fundamentals of wrist-worn accelerometer data, followed by various methods of processing these data (e.g., cut points, steps per minute, machine learning), and then we discuss the opportunities, challenges, and directions for future studies in this area of inquiry. This is the most comprehensive review paper to date regarding the analysis and interpretation of free-living physical activity data derived from wrist-worn accelerometers, aiming to help establish a blueprint for processing wrist-derived accelerometer data.

Examining 24-Hour Activity and Sleep Behaviors and Related Determinants in Latino Adolescents and Young Adults With Obesity

Background. Few studies have examined 24-hour activity and sleep behaviors and their contribution to type 2 diabetes (T2D) in Latino adolescents and young adults with obesity. Aim. This study included quantitative data on T2D risk and 24-hour activity and sleep behaviors and qualitative data on individual, social, and environmental behavioral determinants. Method. A 7 day, 24-hour, wrist-worn accelerometer protocol assessed moderate-to-vigorous physical activity (PA), sedentary behaviors (SB), sleep, and sleep regularity, in adolescents (N = 38; 12-16 years) and young adults (N = 22; 18-22 years). T2D-related outcomes included adiposity (BMI, BF%, waist circumference), fasting, and 2-hour glucose. A subsample of participants (N = 16 adolescents, N = 15 young adults) completed interviews to identify behavioral determinants. Results. High levels of PA were observed among adolescents (M = 103.8 ± 67.5 minutes/day) and young adults (M = 96.8 ± 78.8 minutes/day) as well as high levels of SB across both age groups (≥10 hours/day). Sleep regularity was negatively associated with adiposity (all ps < .05) in both age groups as well as fasting and 2-hour glucose in young adults (all ps < .05). Social support was associated with PA in both age groups as well as SB in younger youth. Auditory noises, lights, and safety inhibited sleep in both age groups. Conclusion. PA is critical for disease reduction, yet reducing SB and improving sleep are also important targets for reducing T2D risk in Hispanic adolescents and young adults. Future health promotion and disease prevention strategies should leverage qualitative findings regarding behavioral determinants.

Human Activity Classification Using Multilayer Perceptron

The number of smart homes is rapidly increasing. Smart homes typically feature functions such as voice-activated functions, automation, monitoring, and tracking events. Besides comfort and convenience, the integration of smart home functionality with data processing methods can provide valuable information about the well-being of the smart home residence. This study is aimed at taking the data analysis within smart homes beyond occupancy monitoring and fall detection. This work uses a multilayer perceptron neural network to recognize multiple human activities from wrist- and ankle-worn devices. The developed models show very high recognition accuracy across all activity classes. The cross-validation results indicate accuracy levels above 98% across all models, and scoring evaluation methods only resulted in an average accuracy reduction of 10%.

Machine-learning models for activity class prediction: A comparative study of feature selection and classification algorithms

PURPOSE

Machine-learning (ML) approaches have been repeatedly coupled with raw accelerometry to classify physical activity classes, but the features required to optimize their predictive performance are still unknown. Our aim was to identify appropriate combination of feature subsets and prediction algorithms for activity class prediction from hip-based raw acceleration data.

METHODS

The hip-based raw acceleration data collected from 27 participants was split into training (70 %) and validation (30 %) subsets. A total of 206 time- (TD) and frequencydomain (FD) features were extracted from 6-second non-overlapping windows of the signal. Feature selection was done using seven filter-based, two wrapper-based, and one embedded algorithm, and classification was performed with artificial neural network (ANN), support vector machine (SVM), and random forest (RF). For every combination between the feature selection method and the classifiers, the most appropriate feature subsets were found and used for model training within the training set. These models were then validated with the left-out validation set.

RESULTS

The appropriate number of features for the ANN, SVM, and RF ranged from 20 to 45. Overall, the accuracy of all the three classifiers was higher when trained with feature subsets generated using filter-based methods compared with when they were trained with wrapper-based methods (range: 78.1 %-88 % vs. 66 %-83.5 %). TD features that reflect how signals vary around the mean, how they differ with one another, and how much and how often they change were more frequently selected via the feature selection methods.

CONCLUSIONS

A subset of TD features from raw accelerometry could be sufficient for ML-based activity classification if properly selected from different axes.

Deep Learning for Classifying Physical Activities from Accelerometer Data

Physical inactivity increases the risk of many adverse health conditions, including the world’s major non-communicable diseases, such as coronary heart disease, type 2 diabetes, and breast and colon cancers, shortening life expectancy. There are minimal medical care and personal trainers’ methods to monitor a patient’s actual physical activity types. To improve activity monitoring, we propose an artificial-intelligence-based approach to classify physical movement activity patterns. In more detail, we employ two deep learning (DL) methods, namely a deep feed-forward neural network (DNN) and a deep recurrent neural network (RNN) for this purpose. We evaluate the two models on two physical movement datasets collected from several volunteers who carried tri-axial accelerometer sensors. The first dataset is from the UCI machine learning repository, which contains 14 different activities-of-daily-life (ADL) and is collected from 16 volunteers who carried a single wrist-worn tri-axial accelerometer. The second dataset includes ten other ADLs and is gathered from eight volunteers who placed the sensors on their hips. Our experiment results show that the RNN model provides accurate performance compared to the state-of-the-art methods in classifying the fundamental movement patterns with an overall accuracy of 84.89% and an overall F1-score of 82.56%. The results indicate that our method provides the medical doctors and trainers a promising way to track and understand a patient’s physical activities precisely for better treatment.

Individual versus Group Calibration of Machine Learning Models for Physical Activity Assessment Using Body-Worn Accelerometers

Modeling approaches for translating accelerometer data into physical activity metrics are often developed using a group calibration approach. However, it is unknown if models developed for specific individuals will improve measurement accuracy.

PURPOSE

We sought to determine if individually calibrated machine learning models yielded higher accuracy than a group calibration approach for physical activity intensity assessment.

METHODS

Participants (n = 48) wore accelerometers on the right hip and non-dominant wrist while performing activities of daily living in a semi-structured laboratory and/or free-living setting. Criterion measures of activity intensity (sedentary, light, moderate, vigorous) were determined using direct observation. Data were reintegrated into 30-second epochs, and eight random forest models were created to determine physical activity intensity by using all possible conditions of training data (individual vs. group), protocol (laboratory vs. free-living), and placement (hip vs. wrist). A 2x2x2 repeated-measures analysis of variance was used to compare epoch-level accuracy statistics (% accuracy, kappa [k]) of the models when used to determine activity intensity in an independent sample of free-living participants.

RESULTS

Main effects were significant for the type of training data (group: accuracy = 80%, k = 0.59; individual: accuracy = 74% [p = 0.02], k = 0.50 [p = 0.01]) and protocol (free-living: accuracy = 81%, k = 0.63; laboratory: accuracy = 74% [p = 0.04], k = 0.47 [p < 0.01]). Main effects were not significant for placement (hip: accuracy = 79%, k = 0.58; wrist: accuracy = 75% [p = 0.18]; k = 0.52 [p = 0.18]). Point estimates for mean absolute error were generally lowest for the group training, free-living protocol, and hip placement.

CONCLUSION

Contrary to expectations, individually calibrated machine learning models yielded poorer accuracy than a traditional group approach. Additionally, models should be developed in free-living settings when possible to optimize predictive accuracy.

Children's Activity Classification for Domestic Risk Scenarios Using Environmental Sound and a Bayesian Network

Children’s healthcare is a relevant issue, especially the prevention of domestic accidents, since it has even been defined as a global health problem. Children’s activity classification generally uses sensors embedded in children’s clothing, which can lead to erroneous measurements for possible damage or mishandling. Having a non-invasive data source for a children’s activity classification model provides reliability to the monitoring system where it is applied. This work proposes the use of environmental sound as a data source for the generation of children’s activity classification models, implementing feature selection methods and classification techniques based on Bayesian networks, focused on the recognition of potentially triggering activities of domestic accidents, applicable in child monitoring systems. Two feature selection techniques were used: the Akaike criterion and genetic algorithms. Likewise, models were generated using three classifiers: naive Bayes, semi-naive Bayes and tree-augmented naive Bayes. The generated models, combining the methods of feature selection and the classifiers used, present accuracy of greater than 97% for most of them, with which we can conclude the efficiency of the proposal of the present work in the recognition of potentially detonating activities of domestic accidents.

Comparison of machine learning techniques for the identification of human activities from inertial sensors available in a mobile device after the application of data imputation techniques

Human activity recognition (HAR) is a significant research area due to its wide range of applications in intelligent health systems, security, and entertainment games. Over the past few years, many studies have recognized human daily living activities using different machine learning approaches. However, the performance of a machine learning algorithm varies based on the sensing device type, the number of sensors in that device, and the position of the underlying sensing device. Moreover, the incomplete activities (i.e., data captures) in a dataset also play a crucial role in the performance of machine learning algorithms. Therefore, we perform a comparative analysis of eight commonly used machine learning algorithms in different sensor combinations in this work. We used a publicly available mobile sensors dataset and applied the k-Nearest Neighbors (KNN) data imputation technique for extrapolating the missing samples. Afterward, we performed a couple of experiments to figure out which algorithm performs best at which sensors' data combination. The experimental analysis reveals that the AdaBoost algorithm outperformed all machine learning algorithms for recognizing five different human daily living activities with both single and multi-sensor combinations. Furthermore, the experimental results show that AdaBoost is capable to correctly identify all the activities presented in the dataset with 100% classification accuracy.

Assessment of Physical Activity in Adults using Wrist Accelerometers

The health benefits of physical activity have been widely recognized, yet traditional measures of physical activity including questionnaires and category-based assessments of volume and intensity provide only broad estimates of daily activities. Accelerometers have advanced epidemiologic research on physical activity by providing objective and continuous measurement of physical activity in free-living conditions. Wrist-worn accelerometers have become especially popular due to low participant burden. However, the validity and reliability of wrist-worn devices for adults have yet to be summarized. Moreover, accelerometer data provide rich information on how physical activity is accumulated throughout the day, but only a small portion of these rich data have been utilized by researchers. Lastly, new methodological developments that aim to overcome some of the limitations of accelerometers are emerging. The purpose of this review is to provide an overview of accelerometry research, with a special focus on wrist-worn accelerometers. We describe briefly how accelerometers work, summarize the validity and reliability of wrist-worn accelerometers, discuss the benefits of accelerometers including measuring light-intensity physical activity, and discuss pattern metrics of daily physical activity recently introduced in the literature. A summary of large-scale cohort studies and randomized trials that implemented wrist-worn accelerometry is provided. We conclude the review by discussing new developments and future directions of research using accelerometers, with a focus on wrist-worn accelerometers.

Adaptive Accumulation of Plantar Pressure for Ambulatory Activity Recognition and Pedestrian Identification

In this paper, we propose a novel method for ambulatory activity recognition and pedestrian identification based on temporally adaptive weighting accumulation-based features extracted from categorical plantar pressure. The method relies on three pressure-related features, which are calculated by accumulating the pressure of the standing foot in each step over three different temporal weighting forms. In addition, we consider a feature reflecting the pressure variation. These four features characterize the standing posture in a step by differently weighting step pressure data over time. We use these features to analyze the standing foot during walking and then recognize ambulatory activities and identify pedestrians based on multilayer multiclass support vector machine classifiers. Experimental results show that the proposed method achieves 97% accuracy for the two tasks when analyzing eight consecutive steps. For faster processing, the method reaches 89.9% and 91.3% accuracy for ambulatory activity recognition and pedestrian identification considering two consecutive steps, respectively, whereas the accuracy drops to 83.3% and 82.3% when considering one step for the respective tasks. Comparative results demonstrated the high performance of the proposed method regarding accuracy and temporal sensitivity.

Development and validation of smartwatch-based activity recognition models for rigging crew workers on cable logging operations

Analysis of high-resolution inertial sensor and global navigation satellite system (GNSS) data collected by mobile and wearable devices is a relatively new methodology in forestry and safety research that provides opportunities for modeling work activities in greater detail than traditional time study analysis. The objective of this study was to evaluate whether smartwatch-based activity recognition models could quantify the activities of rigging crew workers setting and disconnecting log chokers on cable logging operations. Four productive cycle elements (travel to log, set choker, travel away, clear) were timed for choker setters and four productive cycle elements (travel to log, unhook, travel away, clear) were timed for chasers working at five logging sites in North Idaho. Each worker wore a smartwatch that recorded accelerometer data at 25 Hz. Random forest machine learning was used to develop predictive models that classified the different cycle elements based on features extracted from the smartwatch acceleration data using 15 sliding window sizes (1 to 15 s) and five window overlap levels (0%, 25%, 50%, 75%, and 90%). Models were compared using multiclass area under the Receiver Operating Characteristic (ROC) curve, or AUC. The best choker setter model was created using a 3-s window with 90% overlap and had sensitivity values ranging from 76.95% to 83.59% and precision values ranging from 41.42% to 97.08%. The best chaser model was created using a 1-s window with 90% overlap and had sensitivity values ranging from 71.95% to 82.75% and precision values ranging from 14.74% to 99.16%. These results have demonstrated the feasibility of quantifying forestry work activities using smartwatch-based activity recognition models, a basic step needed to develop real-time safety notifications associated with high-risk job functions and to advance subsequent, comparative analysis of health and safety metrics across stand, site, and work conditions.

Age Differences in Estimating Physical Activity by Wrist Accelerometry Using Machine Learning

Accelerometer-based fitness trackers and smartwatches are proliferating with incessant attention towards health tracking. Despite their growing popularity, accurately measuring hallmark measures of physical activities has yet to be accomplished in adults of all ages. In this work, we evaluated the performance of four machine learning models: decision tree, random forest, extreme gradient boosting (XGBoost) and least absolute shrinkage and selection operator (LASSO), to estimate the hallmark measures of physical activities in young (20-50 years), middle-aged (50-70 years], and older adults (70-89 years]. Our models were built to recognize physical activity types, recognize physical activity intensities, estimate energy expenditure (EE) and recognize individual physical activities using wrist-worn tri-axial accelerometer data (33 activities per participant) from a large sample of participants (n = 253, 62% women, aged 20-89 years old). Results showed that the machine learning models were quite accurate at recognizing physical activity type and intensity and estimating energy expenditure. However, models performed less optimally when recognizing individual physical activities. F1-Scores derived from XGBoost's models were high for sedentary (0.955-0.973), locomotion (0.942-0.964) and lifestyle (0.913-0.949) activity types with no apparent difference across age groups. Low (0.919-0.947), light (0.813-0.828) and moderate (0.846-0.875) physical activity intensities were also recognized accurately. The root mean square error range for EE was approximately 1 equivalent of resting EE [0.835-1.009 METs]. Generally, random forest and XGBoost models outperformed other models. In conclusion, machine learning models to label physical activity types, activity intensity and energy expenditure are accurate and there are minimal differences in their performance across young, middle-aged and older adults.

Enterprise supply chain risk assessment based on improved neural network algorithm and machine learning

It is of practical significance to study the decision-making subject in the supply chain under the influence of risk aversion to make a decision and make the supply chain compete in an orderly market environment. In order to improve the effect of enterprise supply chain risk assessment, this paper improves the traditional neural network algorithm, combines machine learning methods and supply chain risk assessment time requirements to set system function modules, and builds the overall system structure. Considering the multiple relationship attributes of supply chain risk knowledge, this paper uses a multi-element semantic network to represent the network structure of supply chain risk knowledge, and proposes a multi-level inventory control modelThis is based on the inventory of the coordination center and other retailers’ procurement/relocation strategy models. After building the model, this paper designs a simulation test to verify and analyze the model performance. The research results show that the model proposed in this paper has a certain effect.

Grassland resource evaluation based on improved bp network model and analytic hierarchy process

Grassland resources are an important part of land resources. Moreover, it has the functions of regulating the climate, windproof and sand fixation, conserving water sources, maintaining water and soil, raising livestock, providing food, purifying the air, and beautifying the environment in terrestrial ecosystems. Grassland resource evaluation is of great significance to the sustainable development of grassland resources. Therefore, this paper improves the BP neural network, uses the comprehensive index method to calculate the weights in the analytic hierarchy process, and constructs a water resources carrying capacity research and analysis system based on the entropy weight extension decision theory. Meanwhile, this paper analyzes different levels of resource and environmental carrying capacity to achieve the purpose of comprehensive evaluation of resource and environmental carrying capacity. In addition, based on the theory of sustainable development, under the guidance of the principle of index system construction, this paper studies the actual situation of grassland resources and the availability and operability of data, and combines with the opinions given by experts to form an evaluation index system of grassland resources and environmental carrying capacity. Finally, through the actual case study analysis, it is concluded that the model constructed in this paper has a certain effect.

Precision nutrition: A systematic literature review

Precision Nutrition research aims to use personal information about individuals or groups of individuals to deliver nutritional advice that, theoretically, would be more suitable than generic advice. Machine learning, a subbranch of Artificial Intelligence, has promise to aid in the development of predictive models that are suitable for Precision Nutrition. As such, recent research has applied machine learning algorithms, tools, and techniques in precision nutrition for different purposes. However, a systematic overview of the state-of-the-art on the use of machine learning in Precision Nutrition is lacking. Therefore, we carried out a Systematic Literature Review (SLR) to provide an overview of where and how machine learning has been used in Precision Nutrition from various aspects, what such machine learning models use as input features, what the availability status of the data used in the literature is, and how the models are evaluated. Nine research questions were defined in this study. We retrieved 4930 papers from electronic databases and 60 primary studies were selected to respond to the research questions. All of the selected primary studies were also briefly discussed in this article. Our results show that fifteen problems spread across seven domains of nutrition and health are present. Four machine learning tasks are seen in the form of regression, classification, recommendation and clustering, with most of these utilizing a supervised approach. In total, 30 algorithms were used, with 19 appearing more than once. Models were through the use of four groups of approaches and 23 evaluation metrics. Personalized approaches are promising to reduce the burden of these current problems in nutrition research, and the current review shows Machine Learning can be incorporated into Precision Nutrition research with high performance. Precision Nutrition researchers should consider incorporating Machine Learning into their methods to facilitate the integration of many complex features, allowing for the development of high-performance Precision Nutrition approaches.

Physical ACTivity in Survivorship (PACTS): study protocol for a randomized controlled trial evaluating a goal-directed therapeutic exercise program in pediatric posterior fossa brain tumor survivors

BACKGROUND

Posterior fossa brain tumors (PFBT) are the most common solid tumor in children. Recent increases in survival rates are encouraging; however, survivors may experience a plethora of disease- and treatment-related complications that can persist into adulthood. Therapeutic exercise interventions have been shown to improve quality of survivorship in other pediatric cancer diagnoses. There is also evidence that goal-directed interventions are effective at improving motor activities, function, and self-care in children with complex health conditions. Yet, there is currently no evidence on the efficacy of goal-directed therapeutic exercise in pediatric PFBT survivors. The Physical ACTivity in Survivorship (PACTS) study aims to investigate the effects of a novel goal-directed therapeutic exercise program on cardiorespiratory fitness and physical activity-related goal attainment in pediatric survivors of PFBT.

METHOD

PFBT survivors, aged five to 17 years, who underwent surgery at least 12 months earlier and completed radiation therapy and/or chemotherapy at least 6 months prior will be recruited from the Queensland Children's Hospital (Brisbane, Australia) (target n = 48). Following baseline assessment, participants are randomized into either the intervention or usual care group. The intervention group will receive weekly individualized, goal-directed exercise therapy delivered face-to-face for 12 weeks, along with an accompanying home-based program (three sessions per week). Outcomes will be assessed at baseline, immediately post-intervention, and at 6- and 12-months post-intervention. The primary outcomes are cardiorespiratory fitness (Peak VO2) and physical activity-related goal attainment. Secondary outcomes are cardiorespiratory endurance, high-level mobility skills, functional muscle strength, habitual physical activity, gait, balance, quality of life, fatigue, participation, perceived movement skill competence and parameters of body composition.

DISCUSSION

PACTS is the first study to investigate the efficacy of goal-directed therapeutic exercise in children with PFBT and provide evidence needed to inform clinical practice recommendations for managing quality of survivorship in PFBT survivors.

TRIAL REGISTRATION

ACTRN12619000841178 .

Free-living Evaluation of Laboratory-based Activity Classifiers in Preschoolers

Machine learning classification models for accelerometer data are potentially more accurate methods to measure physical activity in young children than traditional cut point methods. However, existing algorithms have been trained on laboratory-based activity trials, and their performance has not been investigated under free-living conditions.

PURPOSE

This study aimed to evaluate the accuracy of laboratory-trained hip and wrist random forest and support vector machine classifiers for the automatic recognition of five activity classes: sedentary (SED), light-intensity activities and games (LIGHT_AG), walking (WALK), running (RUN), and moderate to vigorous activities and games (MV_AG) in preschool-age children under free-living conditions.

METHODS

Thirty-one children (4.0 ± 0.9 yr) were video recorded during a 20-min free-living play session while wearing an ActiGraph GT3X+ on their right hip and nondominant wrist. Direct observation was used to continuously code ground truth activity class and specific activity types occurring within each class using a bespoke two-stage coding scheme. Performance was assessed by calculating overall classification accuracy and extended confusion matrices summarizing class-level accuracy and the frequency of specific activities observed within each class.

RESULTS

Accuracy values for the hip and wrist random forest algorithms were 69.4% and 59.1%, respectively. Accuracy values for hip and wrist support vector machine algorithms were 66.4% and 59.3%, respectively. Compared with the laboratory cross validation, accuracy decreased by 11%-15% for the hip classifiers and 19%-21% for the wrist classifiers. Classification accuracy values were 72%-78% for SED, 58%-79% for LIGHT_AG, 71%-84% for MV_AG, 9%-15% for WALK, and 66%-75% for RUN.

CONCLUSION

The accuracy of laboratory-based activity classifiers for preschool-age children was attenuated when tested on new data collected under free-living conditions. Future studies should train and test machine learning activity recognition algorithms using accelerometer data collected under free-living conditions.

Random forest algorithms for recognizing daily life activities using plantar pressure information: a smart-shoe study

Background

Wearable activity trackers are regarded as a new opportunity to deliver health promotion interventions. Indeed, while the prediction of active behaviors is currently primarily relying on the processing of accelerometer sensor data, the emergence of smart clothes with multi-sensing capacities is offering new possibilities. Algorithms able to process data from a variety of smart devices and classify daily life activities could therefore be of particular importance to achieve a more accurate evaluation of physical behaviors. This study aims to (1) develop an activity recognition algorithm based on the processing of plantar pressure information provided by a smart-shoe prototype and (2) to determine the optimal hardware and software configurations.

Method

Seventeen subjects wore a pair of smart-shoe prototypes composed of plantar pressure measurement insoles, and they performed the following nine activities: sitting, standing, walking on a flat surface, walking upstairs, walking downstairs, walking up a slope, running, cycling, and completing office work. The insole featured seven pressure sensors. For each activity, at least four minutes of plantar pressure data were collected. The plantar pressure data were cut in overlapping windows of different lengths and 167 features were extracted for each window. Data were split into training and test samples using a subject-wise assignment method. A random forest model was trained to recognize activity. The resulting activity recognition algorithms were evaluated on the test sample. A multi hold-out procedure allowed repeating the operation with 5 different assignments. The analytic conditions were modulated to test (1) different window lengths (1–60 seconds), (2) some selected sensor configurations and (3) different numbers of data features.

Results

A window length of 20 s was found to be optimum and therefore used for the rest of the analysis. Using all the sensors and all 167 features, the smart shoes predicted the activities with an average success of 89%. “Running” demonstrated the highest sensitivity (100%). “Walking up a slope” was linked with the lowest performance (63%), with the majority of the false negatives being “walking on a flat surface” and “walking upstairs.” Some 2- and 3-sensor configurations were linked with an average success rate of 87%. Reducing the number of features down to 20 does not alter significantly the performance of the algorithm.

Conclusion

High-performance human behavior recognition using plantar pressure data only is possible. In the future, smart-shoe devices could contribute to the evaluation of daily physical activities. Minimalist configurations integrating only a small number of sensors and computing a reduced number of selected features could maintain a satisfying performance. Future experiments must include a more heterogeneous population.

Multi-level modeling with nonlinear movement metrics to classify self-injurious behaviors in autism spectrum disorder

Self-injurious behavior (SIB) is among the most dangerous concerns in autism spectrum disorder (ASD), often requiring detailed and tedious management methods. Sensor-based behavioral monitoring could address the limitations of these methods, though the complex problem of classifying variable behavior should be addressed first. We aimed to address this need by developing a group-level model accounting for individual variability and potential nonlinear trends in SIB, as a secondary analysis of existing data. Ten participants with ASD and SIB engaged in free play while wearing accelerometers. Movement data were collected from > 200 episodes and 18 different types of SIB. Frequency domain and linear movement variability measures of acceleration signals were extracted to capture differences in behaviors, and metrics of nonlinear movement variability were used to quantify the complexity of SIB. The multi-level logistic regression model, comprising of 12 principal components, explained > 65% of the variance, and classified SIB with > 75% accuracy. Our findings imply that frequency-domain and movement variability metrics can effectively predict SIB. Our modeling approach yielded superior accuracy than commonly used classifiers (~ 75 vs. ~ 64% accuracy) and had superior performance compared to prior reports (~ 75 vs. ~ 69% accuracy) This work provides an approach to generating an accurate and interpretable group-level model for SIB identification, and further supports the feasibility of developing a real-time SIB monitoring system.

Classification of human physical activity based on raw accelerometry data via spherical coordinate transformation

Human health is strongly associated with person's lifestyle and levels of physical activity. Therefore, characterization of daily human activity is an important task. Accelerometers have been used to obtain precise measurements of body acceleration. Wearable accelerometers collect data as a three-dimensional time series with frequencies up to 100 Hz. Using such accelerometry signal, we are able to classify different types of physical activity. In our work, we present a novel procedure for physical activity classification based on the raw accelerometry signal. Our proposal is based on the spherical representation of the data. We classify four activity types: resting, upper body activities (sitting), upper body activities (standing), and lower body activities. The classifier is constructed using decision trees with extracted features consisting of spherical coordinates summary statistics, moving averages of the radius and the angles, radius variance, and spherical variance. The classification accuracy of our method has been tested on data collected on a sample of 47 elderly individuals who performed a series of activities in laboratory settings. The achieved classification accuracy is over 90% when the subject-specific data are used and 84% when the group data are used. Main contributor to the classification accuracy is the angular part of the collected signal, especially spherical variance. To the best of our knowledge, spherical variance has never been previously used in the analysis of the raw accelerometry data. Its major advantage over other angular measures is its invariance to the accelerometer location shifts.

Machine learning to quantify habitual physical activity in children with cerebral palsy

AIM

To investigate whether activity-monitors and machine learning models could provide accurate information about physical activity performed by children and adolescents with cerebral palsy (CP) who use mobility aids for ambulation.

METHOD

Eleven participants (mean age 11y [SD 3y]; six females, five males) classified in Gross Motor Function Classification System (GMFCS) levels III and IV, completed six physical activity trials wearing a tri-axial accelerometer on the wrist, hip, and thigh. Trials included supine rest, upper-limb task, walking, wheelchair propulsion, and cycling. Three supervised learning algorithms (decision tree, support vector machine [SVM], random forest) were trained on features in the raw-acceleration signal. Model-performance was evaluated using leave-one-subject-out cross-validation accuracy.

RESULTS

Cross-validation accuracy for the single-placement models ranged from 59% to 79%, with the best performance achieved by the random forest wrist model (79%). Combining features from two or more accelerometer placements significantly improved classification accuracy. The random forest wrist and hip model achieved an overall accuracy of 92%, while the SVM wrist, hip, and thigh model achieved an overall accuracy of 90%.

INTERPRETATION

Models trained on features in the raw-acceleration signal may provide accurate recognition of clinically relevant physical activity behaviours in children and adolescents with CP who use mobility aids for ambulation in a controlled setting.

WHAT THIS PAPER ADDS

Machine learning may assist clinicians in evaluating the efficacy of surgical and therapy-based interventions. Machine learning may help researchers better understand the short- and long-term benefits of physical activity for children with more severe motor impairments.

Indirect Recognition of Predefined Human Activities

The work investigates the application of artificial neural networks and logistic regression for the recognition of activities performed by room occupants. KNX (Konnex) standard-based devices were selected for smart home automation and data collection. The obtained data from these devices (Humidity, CO₂, temperature) were used in combination with two wearable gadgets to classify specific activities performed by the room occupant. The obtained classifications can benefit the occupant by monitoring the wellbeing of elderly residents and providing optimal air quality and temperature by utilizing heating, ventilation, and air conditioning control. The obtained results yield accurate classification.

The AgingPLUS trial: Design of a randomized controlled trial to increase physical activity in middle-aged and older adults

BACKGROUND

Negative views of aging (NVOA), low self-efficacy beliefs, and poor goal planning skills represent risk factors that undermine adults' motivation to engage in physical activity (PA). Targeting these three risk factors may motivate adults to become physically active.

OBJECTIVE

To assess the efficacy of Aging^PLUS, a 4-week educational program that explicitly targets NVOA, low self-efficacy beliefs, and poor goal planning skills compared to a 4-week health education program. The study also examines the role of NVOA, self-efficacy beliefs, and goal planning as the mechanisms underlying change in PA.

DESIGN

This randomized controlled trial (RCT) utilizes the experimental medicine approach to assess change in PA as a function of modifying three risk factors. The RCT recruitment target includes 288 mostly sedentary adults ranging in age from 45 to 75 years.

METHODS

Eligible middle-aged and older adults are recruited through community sources. Participants are randomized to either the Aging^PLUS or the control group. Participants in both groups are enrolled in the trial for 8 months total, with four assessment points: Baseline (pre-test), Week 4 (immediate post-test), Week 8 (delayed post-test), and Month 6 (long-term follow-up). The intervention takes place over 4 consecutive weeks with 2-h sessions each week. PA engagement is the primary outcome variable. Positive changes in NVOA, self-efficacy beliefs, and goal planning are the intervention targets and hypothesized mediators of increases in PA.

SUMMARY

By utilizing a multi-component approach and targeting a cluster of psychological mechanisms, the Aging^PLUS program implements the experimental medicine approach to health behavior change.

Machine Learning Models for Classifying Physical Activity in Free-Living Preschool Children

Machine learning (ML) activity classification models trained on laboratory-based activity trials exhibit low accuracy under free-living conditions. Training new models on free-living accelerometer data, reducing the number of prediction windows comprised of multiple activity types by using shorter windows, including temporal features such as standard deviation in lag and lead windows, and using multiple sensors may improve the classification accuracy under free-living conditions. The objective of this study was to evaluate the accuracy of Random Forest (RF) activity classification models for preschool-aged children trained on free-living accelerometer data. Thirty-one children (mean age = 4.0 ± 0.9 years) completed a 20 min free-play session while wearing an accelerometer on their right hip and non-dominant wrist. Video-based direct observation was used to categorize the children’s movement behaviors into five activity classes. The models were trained using prediction windows of 1, 5, 10, and 15 s, with and without temporal features. The models were evaluated using leave-one-subject-out-cross-validation. The F-scores improved as the window size increased from 1 to 15 s (62.6%–86.4%), with only minimal improvements beyond the 10 s windows. The inclusion of temporal features increased the accuracy, mainly for the wrist classification models, by an average of 6.2 percentage points. The hip and combined hip and wrist classification models provided comparable accuracy; however, both the models outperformed the models trained on wrist data by 7.9 to 8.2 percentage points. RF activity classification models trained with free-living accelerometer data provide accurate recognition of young children’s movement behaviors under real-world conditions.

Simple Method for the Objective Activity Type Assessment with Preschoolers, Children and Adolescents

Background: The objective and accurate assessment of children's sedentary and physical behavior is important for investigating their relation to health. The purpose of this study is to validate a simple and robust method for the identification of sitting, standing, walking, running and biking performed by preschool children, children and adolescents in the age from 3 to 16 years from a single thigh-worn accelerometer. Method: A total of 96 children were included in the study and all subjects followed a structured activity protocol performed in the subject's normal kindergarten or school environment. Thigh acceleration was measured using the Axivity AX3 (Axivity, Newcastle, UK) device. Method development and accuracy was evaluated by equally dividing the subjects into a development and test group. Results: The sensitivity and specificity for identifying sitting and standing was above 99.3% and for walking and running above 82.6% for all age groups. The sensitivity and specificity for identifying biking was above 85.8% for children and adolescents and above 64.8% for the preschool group using running bikes. Conclusion: The accurate assessment of sitting, standing, walking, running and biking from thigh acceleration and with children in the age range of 3 to 16 is valid, although not with preschool children using running bikes.

Detecting and Classifying Self-injurious Behavior in Autism Spectrum Disorder Using Machine Learning Techniques

Traditional self-injurious behavior (SIB) management can place compliance demands on the caregiver and have low ecological validity and accuracy. To support an SIB monitoring system for autism spectrum disorder (ASD), we evaluated machine learning methods for detecting and distinguishing diverse SIB types. SIB episodes were captured with body-worn accelerometers from children with ASD and SIB. The highest detection accuracy was found with k-nearest neighbors and support vector machines (up to 99.1% for individuals and 94.6% for grouped participants), and classification efficiency was quite high (offline processing at ~ 0.1 ms/observation). Our results provide an initial step toward creating a continuous and objective smart SIB monitoring system, which could in turn facilitate the future care of a pervasive concern in ASD.

Hidden Markov model-based activity recognition for toddlers

OBJECTIVE

Physical activity has been shown to impact future health outcomes in adults, but little is known about the long-term impact of physical activity in toddlers. Accurately measuring the specific types and amounts of physical activity in toddlers will help us to understand, predict, and better affect their future health outcomes. Although activity recognition has been extensively developed for adults as well as older children, toddlers move in ways that are significantly different from older children, indicating the need for a more tailored approach.

APPROACH

In this study, 22 toddlers wore Actigraph waist-worn accelerometers which recorded their movements during guided play. The toddlers were videotaped and their activities were later annotated for the following eight distinct activity classes: lying down, being carried, riding in a stroller, sitting, standing, running/walking, crawling, and climbing up/down. Accelerometer data were extracted in 2 s signal windows and paired with the activities the toddlers were performing during that time interval.

MAIN RESULTS

A variety of classifiers were tuned to a validation set. A random forest classifier was found to achieve the highest accuracy of 63.8% in a test set. To improve the accuracy, a hidden Markov model (HMM) was applied by providing the predictions of the static classifiers as observations. The HMM was able to improve the accuracy to 64.8% with all five classifiers increasing the accuracy an average of 1.3% points (95% confidence interval  =  0.7-1.9, p   <  0.01). When the three most misclassified activities (sitting, standing, and riding in a stroller) were collapsed together, the accuracy increased to 79.3%.

SIGNIFICANCE

Further refinement of the toddler activity recognition classifier will enable more accurate measurements of toddler activity and improve future health outcomes of toddlers.

Application of Raw Accelerometer Data and Machine-Learning Techniques to Characterize Human Movement Behavior: A Systematic Scoping Review

BACKGROUND

Application of machine learning for classifying human behavior is increasingly common as access to raw accelerometer data improves. The aims of this scoping review are (1) to examine if machine-learning techniques can accurately identify human activity behaviors from raw accelerometer data and (2) to summarize the practical implications of these machine-learning techniques for future work.

METHODS

Keyword searches were performed in Scopus, Web of Science, and EBSCO databases in 2018. Studies that applied supervised machine-learning techniques to raw accelerometer data and estimated components of physical activity were included. Information on study characteristics, machine-learning techniques, and key study findings were extracted from included studies.

RESULTS

Of the 53 studies included in the review, 75% were published in the last 5 years. Most studies predicted postures and activity type, rather than intensity, and were conducted in controlled environments using 1 or 2 devices. The most common models were support vector machine, random forest, and artificial neural network. Overall, classification accuracy ranged from 62% to 99.8%, although nearly 80% of studies achieved an overall accuracy above 85%.

CONCLUSIONS

Machine-learning algorithms demonstrate good accuracy when predicting physical activity components; however, their application to free-living settings is currently uncertain.

Development of a Human Activity Recognition System for Ballet Tasks

Background

Accurate and detailed measurement of a dancer’s training volume is a key requirement to understanding the relationship between a dancer’s pain and training volume. Currently, no system capable of quantifying a dancer’s training volume, with respect to specific movement activities, exists. The application of machine learning models to wearable sensor data for human activity recognition in sport has previously been applied to cricket, tennis and rugby. Thus, the purpose of this study was to develop a human activity recognition system using wearable sensor data to accurately identify key ballet movements (jumping and lifting the leg). Our primary objective was to determine if machine learning can accurately identify key ballet movements during dance training. The secondary objective was to determine the influence of the location and number of sensors on accuracy.

Results

Convolutional neural networks were applied to develop two models for every combination of six sensors (6, 5, 4, 3, etc.) with and without the inclusion of transition movements. At the first level of classification, including data from all sensors, without transitions, the model performed with 97.8% accuracy. The degree of accuracy reduced at the second (83.0%) and third (75.1%) levels of classification. The degree of accuracy reduced with inclusion of transitions, reduction in the number of sensors and various sensor combinations.

Conclusion

The models developed were robust enough to identify jumping and leg lifting tasks in real-world exposures in dancers. The system provides a novel method for measuring dancer training volume through quantification of specific movement tasks. Such a system can be used to further understand the relationship between dancers’ pain and training volume and for athlete monitoring systems. Further, this provides a proof of concept which can be easily translated to other lower limb dominant sporting activities

Physical Activity Promotion for Apprentices in Nursing Care and Automotive Mechatronics-Competence Counts More than Volume

Apprentices in the area of nursing care and automotive mechatronics are exposed to increased health risks. In this context, the promotion of physical activity (PA) is considered an effective strategy for the assurance of work ability. The goal of the PArC-AVE study was therefore to better understand the role of PA for apprentices employed in these two sectors. In an exploratory study, 55 apprentices wore an ActiGraph accelerometer over seven consecutive days and were subject to activity analysis. The objective accelerometer data (18,979 ± 3780 steps/day; 471.00 ± 159.75 min of moderate-to-vigorous PA/week), complemented by questionnaire data, indicated that most met the volume-based PA recommendations. Subsequently, we conducted a multicenter study comprising 745 apprentices from six vocational education institutions. Path analyses showed that competencies for health-enhancing PA were significantly related to indicators of work ability (0.180 ≤ b ≤ 0.452) and psychophysical health (0.139 ≤ b ≤ 0.347), whereas mere volume of PA was not (-0.048 ≤ b ≤ 0.080). In summary, apprentices of nursing care and automotive mechatronics showed high levels of PA. However, the results highlight the importance of competencies for health-enhancing PA. The PAHCO model could provide a useful framework for the conceptualization of effective interventions.

Do we have to reduce the recall period? Validity of a daily physical activity questionnaire (PAQ24) in young active adults

Background

Combining the strengths of physical activity (PA) diaries and questionnaires may be needed to improve the unsatisfying measurement quality of existing PA questionnaires. This study investigated the construct validity of a short PA questionnaire (Physical Activity Questionnaire for 24 h [PAQ24]) with a recall period of one day.

Methods

In this cross-sectional study, participants completed the PAQ24 on seven consecutive days while wearing an accelerometer (GENEActiv). Thereafter, the Global Physical Activity Questionnaire (GPAQ) was completed. Spearman correlation coefficients and Bland-Altman analysis were used to assess construct validity.

Results

Overall, 50 active adults (11 women, mean age = 25.1 ± 2.5) participated. Relative agreements between Total PA of PAQ24 and accelerometer were 0.37 ≤ ρ ≤ 0.72 for each day with satisfying agreement on five out of seven days. Weekly relative agreement for Total PA was moderate (ρ = 0.44). Relative agreements between PAQ24 and GPAQ were ρ = 0.43 for Total PA. Daily and weekly absolute agreements were poor indicated by wide limits of agreement.

Conclusions

In contrast to weekly Total PA, the majority of daily results of the PAQ24 showed satisfying construct validity. A short recall period may improve the measurement quality of PA questionnaires, but measurement errors and the costs of multiple administrations must be considered in future studies.