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

Joint association of genetic risk and accelerometer-measured physical activity with incident coronary artery disease in the UK biobank cohort

Previous research demonstrates the joint association of self-reported physical activity and genotype with coronary artery disease. However, an existing research gap is whether accelerometer-measured overall physical activity or physical activity intensity can offset genetic predisposition to coronary artery disease. This study explores the independent and joint associations of accelerometer-measured physical activity and genetic predisposition with incident coronary artery disease. Incident coronary artery disease based on hospital inpatient records and death register data serves as the outcome of this study. Polygenic risk score and overall physical activity, measured as Euclidean Norm Minus One, and intensity, measured as minutes per day of moderate-to-vigorous intensity physical activity (MVPA), are examined both linearly and by decile. The UK Biobank population-based cohort recruited over 500,000 individuals aged 40 to 69 between 2006 and 2010, with 103,712 volunteers participating in a weeklong wrist-worn accelerometer study from 2013 to 2015. Individuals of White British ancestry (n = 65,079) meeting the genotyping and accelerometer-based inclusion criteria and with no missing covariates were included in the analytic sample. In the sample of 65,079 individuals, the mean (SD) age was 62.51 (7.76) and 61% were female. During a median follow-up of 6.8 years, 1,382 cases of coronary artery disease developed. At the same genetic risk, physical activity intensity had a hazard ratio (HR) of 0.41 (95% CI: 0.29-0.60) at the 90th compared to 10th percentile, equivalent to 31.68 and 120.96 minutes of moderate-to-vigorous physical activity per day, respectively, versus an HR of 0.61 (95% CI: 0.52-0.72) for overall physical activity. The combination of high genetic risk and low physical activity intensity showed the greatest risk, with an individual at the 10th percentile of genetic risk and 90th percentile of intensity facing an HR of 0.14 (95% CI: 0.09-0.21) compared to an individual at the 90th percentile of genetic risk and 10th percentile of intensity. Physical activity, especially physical activity intensity, is associated with an attenuation of some of the risk of coronary artery disease but this pattern does not vary by genetic risk. This accelerometer-based study provides the clearest evidence to date regarding the joint influence of genetics, overall physical activity, and physical activity intensity on coronary artery disease.

Using functional principal component analysis (FPCA) to quantify sitting patterns derived from wearable sensors

BACKGROUND

Sedentary behavior (SB) is a recognized risk factor for many chronic diseases. ActiGraph and activPAL are two commonly used wearable accelerometers in SB research. The former measures body movement and the latter measures body posture. The goal of the current study is to quantify the pattern and variation of movement (by ActiGraph activity counts) during activPAL-identified sitting events, and examine associations between patterns and health-related outcomes, such as systolic and diastolic blood pressure (SBP and DBP).

METHODS

The current study included 314 overweight postmenopausal women, who were instructed to wear an activPAL (at thigh) and ActiGraph (at waist) simultaneously for 24 hours a day for a week under free-living conditions. ActiGraph and activPAL data were processed to obtain minute-level time-series outputs. Multilevel functional principal component analysis (MFPCA) was applied to minute-level ActiGraph activity counts within activPAL-identified sitting bouts to investigate variation in movement while sitting across subjects and days. The multilevel approach accounted for the nesting of days within subjects.

RESULTS

At least 90% of the overall variation of activity counts was explained by two subject-level principal components (PC) and six day-level PCs, hence dramatically reducing the dimensions from the original minute-level scale. The first subject-level PC captured patterns of fluctuation in movement during sitting, whereas the second subject-level PC delineated variation in movement during different lengths of sitting bouts: shorter (< 30 minutes), medium (30 -39 minutes) or longer (> 39 minute). The first subject-level PC scores showed positive association with DBP (standardized β ^ : 2.041, standard error: 0.607, adjusted p = 0.007), which implied that lower activity counts (during sitting) were associated with higher DBP.

CONCLUSION

In this work we implemented MFPCA to identify variation in movement patterns during sitting bouts, and showed that these patterns were associated with cardiovascular health. Unlike existing methods, MFPCA does not require pre-specified cut-points to define activity intensity, and thus offers a novel powerful statistical tool to elucidate variation in SB patterns and health.

TRIAL REGISTRATION

ClinicalTrials.gov NCT03473145; Registered 22 March 2018; https://clinicaltrials.gov/ct2/show/NCT03473145 ; International Registered Report Identifier (IRRID): DERR1-10.2196/28684.

Classification of Free-Living Body Posture with ECG Patch Accelerometers: Application to the Multicenter AIDS Cohort Study

Purpose

As health studies increasingly monitor free-living heart performance via ECG patches with accelerometers, researchers will seek to investigate cardio-electrical responses to physical activity and sedentary behavior, increasing demand for fast, scalable methods to process accelerometer data. We extend a posture classification algorithm for accelerometers in ECG patches when researchers do not have ground-truth labels or other reference measurements (i.e., upright measurement).

Methods

Men living with and without HIV in the Multicenter AIDS Cohort study wore the Zio XT® for up to two weeks (n = 1,250). Our novel extensions for posture classification include (1) estimation of an upright posture for each individual without a reference upright measurement; (2) correction of the upright estimate for device removal and re-positioning using novel spherical change-point detection; and (3) classification of upright and recumbent periods using a clustering and voting process rather than a simple inclination threshold used in other algorithms. As no posture labels exist in the free-living environment, we perform numerous sensitivity analyses and evaluate the algorithm against labelled data from the Towson Accelerometer Study, where participants wore accelerometers at the waist.

Results

On average, 87.1% of participants were recumbent at 4am and 15.5% were recumbent at 1pm. Participants were recumbent 54 minutes longer on weekends compared to weekdays. Performance was good in comparison to labelled data in a separate, controlled setting (accuracy = 96.0%, sensitivity = 97.5%, specificity = 95.9%).

Conclusions

Posture may be classified in the free-living environment from accelerometers in ECG patches even without measuring a standard upright position. Furthermore, algorithms that fail to account for individuals who rotate and re-attach the accelerometer may fail in the free-living environment.

Development of a multi-wear-site, deep learning-based physical activity intensity classification algorithm using raw acceleration data

BACKGROUND

Accelerometers are widely adopted in research and consumer devices as a tool to measure physical activity. However, existing algorithms used to estimate activity intensity are wear-site-specific. Non-compliance to wear instructions may lead to misspecifications. In this study, we developed deep neural network models to classify device placement and activity intensity based on raw acceleration data. Performances of these models were evaluated by making comparisons to the ground truth and results derived from existing count-based algorithms.

METHODS

54 participants (26 adults 26.9±8.7 years; 28 children, 12.1±2.3 years) completed a series of activity tasks in a laboratory with accelerometers attached to each of their hip, wrist, and chest. Their metabolic rates at rest and during activity periods were measured using the portable COSMED K5; data were then converted to metabolic equivalents, and used as the ground truth for activity intensity. Deep neutral networks using the Long Short-Term Memory approach were trained and evaluated based on raw acceleration data collected from accelerometers. Models to classify wear-site and activity intensity, respectively, were evaluated.

RESULTS

The trained models correctly classified wear-sites and activity intensities over 90% of the time, which outperformed count-based algorithms (wear-site correctly specified: 83% to 85%; wear-site misspecified: 64% to 75%). When additional parameters of age, height and weight of participants were specified, the accuracy of some prediction models surpassed 95%.

CONCLUSIONS

Results of the study suggest that accelerometer placement could be determined prospectively, and non-wear-site-specific algorithms had satisfactory accuracies. The performances, in terms of intensity classification, of these models also exceeded typical count-based algorithms. Without being restricted to one specific wear-site, research protocols for accelerometers wear could allow more autonomy to participants, which may in turn improve their acceptance and compliance to wear protocols, and in turn more accurate results.

Emerging methods for measuring physical activity using accelerometry in children and adolescents with neuromotor disorders: a narrative review

BACKGROUND

Children and adolescents with neuromotor disorders need regular physical activity to maintain optimal health and functional independence throughout their development. To this end, reliable measures of physical activity are integral to both assessing habitual physical activity and testing the efficacy of the many interventions designed to increase physical activity in these children. Wearable accelerometers have been used for children with neuromotor disorders for decades; however, studies most often use disorder-specific cut points to categorize physical activity intensity, which lack generalizability to a free-living environment. No reviews of accelerometer data processing methods have discussed the novel use of machine learning techniques for monitoring physical activity in children with neuromotor disorders.

METHODS

In this narrative review, we discuss traditional measures of physical activity (including questionnaires and objective accelerometry measures), the limitations of standard analysis for accelerometry in this unique population, and the potential benefits of applying machine learning approaches. We also provide recommendations for using machine learning approaches to monitor physical activity.

CONCLUSIONS

While wearable accelerometers provided a much-needed method to quantify physical activity, standard cut point analyses have limitations in children with neuromotor disorders. Machine learning models are a more robust method of analyzing accelerometer data in pediatric neuromotor disorders and using these methods over disorder-specific cut points is likely to improve accuracy of classifying both type and intensity of physical activity. Notably, there remains a critical need for further development of classifiers for children with more severe motor impairments, preschool aged children, and children in hospital settings.

Objective Measurement of Posture and Movement in Young Children Using Wearable Sensors and Customised Mathematical Approaches: A Systematic Review

Given the importance of young children's postures and movements to health and development, robust objective measures are required to provide high-quality evidence. This study aimed to systematically review the available evidence for objective measurement of young (0-5 years) children's posture and movement using machine learning and other algorithm methods on accelerometer data. From 1663 papers, a total of 20 papers reporting on 18 studies met the inclusion criteria. Papers were quality-assessed and data extracted and synthesised on sample, postures and movements identified, sensors used, model development, and accuracy. A common limitation of studies was a poor description of their sample data, yet over half scored adequate/good on their overall study design quality assessment. There was great diversity in all aspects examined, with evidence of increasing sophistication in approaches used over time. Model accuracy varied greatly, but for a range of postures and movements, models developed on a reasonable-sized (n > 25) sample were able to achieve an accuracy of >80%. Issues related to model development are discussed and implications for future research outlined. The current evidence suggests the rapidly developing field of machine learning has clear potential to enable the collection of high-quality evidence on the postures and movements of young children.

Machine Learning in Spinal Cord Stimulation for Chronic Pain

Spinal cord stimulation (SCS) is an effective treatment for chronic neuropathic pain. The success of SCS is dependent on candidate selection, response to trialing, and programming optimization. Owing to the subjective nature of these variables, machine learning (ML) offers a powerful tool to augment these processes. Here we explore what work has been done using data analytics and applications of ML in SCS. In addition, we discuss aspects of SCS which have narrowly been influenced by ML and propose the need for further exploration. ML has demonstrated a potential to complement SCS to an extent ranging from assistance with candidate selection to replacing invasive and costly aspects of the surgery. The clinical application of ML in SCS shows promise for improving patient outcomes, reducing costs of treatment, limiting invasiveness, and resulting in a better quality of life for the patient.

AccNet24: A deep learning framework for classifying 24-hour activity behaviours from wrist-worn accelerometer data under free-living environments

OBJECTIVE

Although machine learning techniques have been repeatedly used for activity prediction from wearable devices, accurate classification of 24-hour activity behaviour categories from accelerometry data remains a challenge. We developed and validated a deep learning-based framework for classifying 24-hour activity behaviours from wrist-worn accelerometers.

METHODS

Using an openly available dataset with free-living wrist-based raw accelerometry data from 151 participants (aged 18-91 years), we developed a deep learning framework named AccNet24 to classify 24-hour activity behaviours. First, the acceleration signal (x, y, and z-axes) was segmented into 30-second nonoverlapping windows, and signal-to-image conversion was performed for each segment. Deep features were automatically extracted from the signal images using transfer learning and transformed into a lower-dimensional feature space. These transformed features were then employed to classify the activity behaviours as sleep, sedentary behaviour, and light-intensity (LPA) and moderate-to-vigorous physical activity (MVPA) using a bidirectional long short-term memory (BiLSTM) recurrent neural network. AccNet24 was trained and validated with data from 101 and 25 randomly selected participants and tested with the remaining unseen 25 participants. We also extracted 112 hand-crafted time and frequency domain features from 30-second windows and used them as inputs to five commonly used machine learning classifiers, including random forest, support vector machines, artificial neural networks, decision tree, and naïve Bayes to classify the 24-hour activity behaviour categories.

RESULTS

Using the same training, validation, and test data and window size, the classification accuracy of AccNet24 outperformed the accuracy of the other five machine learning classification algorithms by 16%-30% on unseen data.

CONCLUSION

AccNet24, relying on signal-to-image conversion, deep feature extraction, and BiLSTM achieved consistently high accuracy (>95 %) in classifying the 24-hour activity behaviour categories as sleep, sedentary, LPA, and MVPA. The next generation accelerometry analytics may rely on deep learning techniques for activity prediction.

Identification and Quantification of Activities Common to Intensive Care Patients; Development and Validation of a Dual-Accelerometer-Based Algorithm

The aim of this study was to develop and validate an algorithm that can identify the type, frequency, and duration of activities common to intensive care (IC) patients. Ten healthy participants wore two accelerometers on their chest and leg while performing 14 activities clustered into four protocols (i.e., natural, strict, healthcare provider, and bed cycling). A video served as the reference standard, with two raters classifying the type and duration of all activities. This classification was reliable as intraclass correlations were all above 0.76 except for walking in the healthcare provider protocol, (0.29). The data of four participants were used to develop and optimize the algorithm by adjusting body-segment angles and rest-activity-threshold values based on percentage agreement (%Agr) with the reference. The validity of the algorithm was subsequently assessed using the data from the remaining six participants. %Agr of the algorithm versus the reference standard regarding lying, sitting activities, and transitions was 95%, 74%, and 80%, respectively, for all protocols except transitions with the help of a healthcare provider, which was 14-18%. For bed cycling, %Agr was 57-76%. This study demonstrated that the developed algorithm is suitable for identifying and quantifying activities common for intensive care patients. Knowledge on the (in)activity of these patients and their impact will optimize mobilization.

Effect of Equipment on the Accuracy of Accelerometer-Based Human Activity Recognition in Extreme Environments

A little explored area of human activity recognition (HAR) is in people operating in relation to extreme environments, e.g., mountaineers. In these contexts, the ability to accurately identify activities, alongside other data streams, has the potential to prevent death and serious negative health events to the operators. This study aimed to address this user group and investigate factors associated with the placement, number, and combination of accelerometer sensors. Eight participants (age = 25.0 ± 7 years) wore 17 accelerometers simultaneously during lab-based simulated mountaineering activities, under a range of equipment and loading conditions. Initially, a selection of machine learning techniques was tested. Secondly, a comprehensive analysis of all possible combinations of the 17 accelerometers was performed to identify the optimum number of sensors, and their respective body locations. Finally, the impact of activity-specific equipment on the classifier accuracy was explored. The results demonstrated that the support vector machine (SVM) provided the most accurate classifications of the five machine learning algorithms tested. It was found that two sensors provided the optimum balance between complexity, performance, and user compliance. Sensors located on the hip and right tibia produced the most accurate classification of the simulated activities (96.29%). A significant effect associated with the use of mountaineering boots and a 12 kg rucksack was established.

Machine Learning Models for Weight-Bearing Activity Type Recognition Based on Accelerometry in Postmenopausal Women

Hip-worn triaxial accelerometers are widely used to assess physical activity in terms of energy expenditure. Methods for classification in terms of different types of activity of relevance to the skeleton in populations at risk of osteoporosis are not currently available. This publication aims to assess the accuracy of four machine learning models on binary (standing and walking) and tertiary (standing, walking, and jogging) classification tasks in postmenopausal women. Eighty women performed a shuttle test on an indoor track, of which thirty performed the same test on an indoor treadmill. The raw accelerometer data were pre-processed, converted into eighteen different features and then combined into nine unique feature sets. The four machine learning models were evaluated using three different validation methods. Using the leave-one-out validation method, the highest average accuracy for the binary classification model, 99.61%, was produced by a k-NN Manhattan classifier using a basic statistical feature set. For the tertiary classification model, the highest average accuracy, 94.04%, was produced by a k-NN Manhattan classifier using a feature set that included all 18 features. The methods and classifiers within this study can be applied to accelerometer data to more accurately characterize weight-bearing activity which are important to skeletal health.

Sociodemographic differences in 24-hour time-use behaviours in New Zealand children

BACKGROUND

The time that children spend in physical activity, sedentary behaviour, and sleep each day (i.e., 24-h time-use behaviours), is related to physical and mental health outcomes. Currently, there is no comprehensive evidence on New Zealand school-aged children's 24-h time-use behaviours, adherence to the New Zealand 24-h Movement Guidelines, and how these vary among different sociodemographic groups.

METHODS

This study utilises data from the 8-year wave of the Growing Up in New Zealand longitudinal study. Using two Axivity AX3 accelerometers, children's 24-h time-use behaviours were described from two perspectives: activity intensity and activity type. Compositional data analysis techniques were used to explore the differences in 24-h time-use compositions across various sociodemographic groups.

RESULTS

Children spent on average, 31.1%, 22.3%, 6.8%, and 39.8% of their time in sedentary, light physical activity, moderate-to-vigorous physical activity, and sleep, respectively. However, the daily distribution of time in different activity types was 33.2% sitting, 10.8% standing, 7.3% walking, 0.4% running, and 48.2% lying. Both the activity intensity and activity type compositions varied across groups of child ethnicity, gender, and household income or deprivation. The proportion of children meeting each of the guidelines was 90% for physical activity, 62.5% for sleep, 16% for screen time, and 10.6% for the combined guidelines. Both gender and residence location (i.e., urban vs. rural) were associated with meeting the physical activity guideline, whereas child ethnicity, mother's education and residence location were associated with meeting the screen time guideline. Child ethnicity and mother's education were also significantly associated with the adherence to the combined 24-h Movement Guidelines.

CONCLUSIONS

This study provided comprehensive evidence on how New Zealand children engage in 24-h time-use behaviours, adherence to the New Zealand 24-h Movement Guidelines, and how these behaviours differ across key sociodemographic groups. These findings should be considered in designing future interventions for promoting healthy time-use patterns in New Zealand children.

A theory-based model of cumulative activity

Energy expenditure can be used to examine the health of individuals and the impact of environmental factors on physical activity. One of the more common ways to quantify energy expenditure is to process accelerometer data into some unit of measurement for this expenditure, such as Actigraph activity counts, and bin those measures into physical activity levels. However, accepted thresholds can vary between demographics, and some units of energy measurements do not currently have agreed upon thresholds. We present an approach which computes unique thresholds for each individual, using piecewise exponential functions to model the characteristics of their overall physical activity patterns corresponding to well established sedentary, light, moderate and vigorous activity levels from the literature. Models are fit using existing piecewise fitting techniques and software. Most participants’ activity intensity profile is exceptionally well modeled as piecewise exponential decay. Using this model, we find emergent groupings of participant behavior and categorize individuals into non-vigorous, consistent, moderately active, or extremely active activity intensity profiles. In the supplemental materials, we demonstrate that the parameters of the model correlate with demographics of age, household size, and level of education, inform behavior change under COVID lockdown, and are reasonably robust to signal frequency.

Exploring Orientation Invariant Heuristic Features with Variant Window Length of 1D-CNN-LSTM in Human Activity Recognition

Many studies have explored divergent deep neural networks in human activity recognition (HAR) using a single accelerometer sensor. Multiple types of deep neural networks, such as convolutional neural networks (CNN), long short-term memory (LSTM), or their hybridization (CNN-LSTM), have been implemented. However, the sensor orientation problem poses challenges in HAR, and the length of windows as inputs for the deep neural networks has mostly been adopted arbitrarily. This paper explores the effect of window lengths with orientation invariant heuristic features on the performance of 1D-CNN-LSTM in recognizing six human activities; sitting, lying, walking and running at three different speeds using data from an accelerometer sensor encapsulated into a smartphone. Forty-two participants performed the six mentioned activities by keeping smartphones in their pants pockets with arbitrary orientation. We conducted an inter-participant evaluation using 1D-CNN-LSTM architecture. We found that the average accuracy of the classifier was saturated to 80 ± 8.07% for window lengths greater than 65 using only four selected simple orientation invariant heuristic features. In addition, precision, recall and F1-measure in recognizing stationary activities such as sitting and lying decreased with increment of window length, whereas we encountered an increment in recognizing the non-stationary activities.

Predicting lying, sitting and walking at different intensities using smartphone accelerometers at three different wear locations: hands, pant pockets, backpack

Objective

This study uses machine learning (ML) to develop methods for estimating activity type/intensity using smartphones, to evaluate the accuracy of these models for classifying activity, and to evaluate differences in accuracy between three different wear locations.

Method

Forty-eight participants were recruited to complete a series of activities while carrying Samsung phones in three different locations: backpack, right hand and right pocket. They were asked to sit, lie down, walk and run three Metabolic Equivalent Task (METs), five METs and at seven METs. Raw accelerometer data were collected. We used the R, activity counts package, to calculate activity counts and generated new features based on the raw accelerometer data. We evaluated and compared several ML algorithms; Random Forest (RF), Support Vector Machine, Naïve Bayes, Decision Tree, Linear Discriminant Analysis and k-Nearest Neighbours using the caret package (V.6.0-86). Using the combination of the raw accelerometer data and the computed features leads to high model accuracy.

Results

Using raw accelerometer data, RF models achieved an accuracy of 92.90% for the right pocket location, 89% for the right hand location and 90.8% for the backpack location. Using activity counts, RF models achieved an accuracy of 51.4% for the right pocket location, 48.5% for the right hand location and 52.1% for the backpack location.

Conclusion

Our results suggest that using smartphones to measure physical activity is accurate for estimating activity type/intensity and ML methods, such as RF with feature engineering techniques can accurately classify physical activity intensity levels in laboratory settings.

Evaluation of Wrist Accelerometer Cut-Points for Classifying Physical Activity Intensity in Youth

Background

Wrist worn accelerometers are convenient to wear and provide greater compliance. However, methods to transform the resultant output into predictions of physical activity (PA) intensity have been slow to evolve, with most investigators continuing the practice of applying intensity-based thresholds or cut-points. The current study evaluated the classification accuracy of seven sets of previously published youth-specific cut-points for wrist worn ActiGraph accelerometer data.

Methods

Eighteen children and adolescents [mean age (± SD) 14.6 ± 2.4 years, 10 boys, 8 girls] completed 12 standardized activity trials. During each trial, participants wore an ActiGraph GT3X+ tri-axial accelerometer on the wrist and energy expenditure (Youth METs) was measured directly using the Oxycon Mobile portable calorimetry system. Seven previously published sets of ActiGraph cut-points were evaluated: Crouter regression vertical axis, Crouter regression vector magnitude, Crouter ROC curve vertical axis, Crouter ROC curve vector magnitude, Chandler ROC curve vertical axis, Chandler ROC curve vector magnitude, and Hildebrand ENMO. Classification accuracy was evaluated via weighted Kappa. Confusion matrices were generated to summarize classification accuracy and identify patterns of misclassification.

Results

The cut-points exhibited only moderate agreement with directly measured PA intensity, with Kappa ranging from 0.45 to 0.58. Although the cut-points classified sedentary behavior accurately (> 95%), classification accuracy for the light (3-51%), moderate (12-45%), and vigorous-intensity trials (30-88%) was generally poor. All cut-points underestimated the true intensity of the walking trials, with error rates ranging from 35 to 100%, while the intensity of activity trials requiring significant upper body and/or arm movements was consistently overestimated. The Hildebrand cut-points which serve as the default option in the popular GGIR software package misclassified 30% of the light intensity trials as sedentary and underestimated the intensity of moderate and vigorous intensity trials 75% of the time.

Conclusion

Published ActiGraph cut-points for the wrist, developed specifically for school-aged youth, do not provide acceptable classification accuracy for estimating daily time spent in light, moderate, and vigorous intensity physical activity. The development and deployment of more robust accelerometer data reduction methods such as functional data analysis and machine learning approaches continues to be a research priority.

Why machine learning (ML) has failed physical activity research and how we can improve

Measuring physical activity is a critical issue for our understanding of the health benefits of human movement. Machine learning (ML), using accelerometer data, has become a common way to measure physical activity. ML has failed physical activity measurement research in four important ways. First, as a field, physical activity researchers have not adopted and used principles from computer science. Benchmark datasets are common in computer science and allow the direct comparison of different ML approaches. Access to and development of benchmark datasets are critical components in advancing ML for physical activity. Second, the priority of methods development focused on ML has created blind spots in physical activity measurement. Methods, other than cut-point approaches, may be sufficient or superior to ML but these are not prioritised in our research. Third, while ML methods are common in published papers, their integration with software is rare. Physical activity researchers must continue developing and integrating ML methods into software to be fully adopted by applied researchers in the discipline. Finally, training continues to limit the uptake of ML in applied physical activity research. We must improve the development, integration and use of software that allows for ML methods’ broad training and application in the field.

CARL: a running recognition algorithm for free-living accelerometer data

Wearable accelerometers hold great promise for physical activity epidemiology and sports biomechanics. However, identifying and extracting data from specific physical activities, such as running, remains challenging.Objective. To develop and validate an algorithm to identify bouts of running in raw, free-living accelerometer data from devices worn at the wrist or torso (waist, hip, chest).Approach. The CARL (continuous amplitude running logistic) classifier identifies acceleration data with amplitude and frequency characteristics consistent with running. The CARL classifier was trained on data from 31 adults wearing accelerometers on the waist and wrist, then validated on free-living data from 30 new, unseen subjects plus 166 subjects from previously-published datasets using different devices, wear locations, and sample frequencies.Main results. On free-living data, the CARL classifier achieved mean accuracy (F₁score) of 0.984 (95% confidence interval 0.962-0.996) for data from the waist and 0.994 (95% CI 0.991-0.996) for data from the wrist. In previously-published datasets, the CARL classifier identified running with mean accuracy (F₁score) of 0.861 (95% CI 0.836-0.884) for data from the chest, 0.911 (95% CI 0.884-0.937) for data from the hip, 0.916 (95% CI 0.877-0.948) for data from the waist, and 0.870 (95% CI 0.834-0.903) for data from the wrist. Misclassification primarily occurred during activities with similar torso acceleration profiles to running, such as rope jumping and elliptical machine use.Significance. The CARL classifier can accurately identify bouts of running as short as three seconds in free-living accelerometry data. An open-source implementation of the CARL classifier is available atgithub.com/johnjdavisiv/carl.

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.

The CNN Hip Accelerometer Posture (CHAP) Method for Classifying Sitting Patterns from Hip Accelerometers: A Validation Study

INTRODUCTION

Sitting patterns predict several healthy aging outcomes. These patterns can potentially be measured using hip-worn accelerometers, but current methods are limited by an inability to detect postural transitions. To overcome these limitations, we developed the Convolutional Neural Network Hip Accelerometer Posture (CHAP) classification method.

METHODS

CHAP was developed on 709 older adults who wore an ActiGraph GT3X+ accelerometer on the hip, with ground-truth sit/stand labels derived from concurrently worn thigh-worn activPAL inclinometers for up to 7 d. The CHAP method was compared with traditional cut-point methods of sitting pattern classification as well as a previous machine-learned algorithm (two-level behavior classification).

RESULTS

For minute-level sitting versus nonsitting classification, CHAP performed better (93% agreement with activPAL) than did other methods (74%-83% agreement). CHAP also outperformed other methods in its sensitivity to detecting sit-to-stand transitions: cut-point (73%), TLBC (26%), and CHAP (83%). CHAP's positive predictive value of capturing sit-to-stand transitions was also superior to other methods: cut-point (30%), TLBC (71%), and CHAP (83%). Day-level sitting pattern metrics, such as mean sitting bout duration, derived from CHAP did not differ significantly from activPAL, whereas other methods did: activPAL (15.4 min of mean sitting bout duration), CHAP (15.7 min), cut-point (9.4 min), and TLBC (49.4 min).

CONCLUSION

CHAP was the most accurate method for classifying sit-to-stand transitions and sitting patterns from free-living hip-worn accelerometer data in older adults. This promotes enhanced analysis of older adult movement data, resulting in more accurate measures of sitting patterns and opening the door for large-scale cohort studies into the effects of sitting patterns on healthy aging outcomes.

Perspective: Opportunities and Challenges of Technology Tools in Dietary and Activity Assessment: Bridging Stakeholder Viewpoints

The science and tools of measuring energy intake and output in humans have rapidly advanced in the last decade. Engineered devices such as wearables and sensors, software applications, and Web-based tools are now ubiquitous in both research and consumer environments. The assessment of energy expenditure in particular has progressed from reliance on self-report instruments to advanced technologies requiring collaboration across multiple disciplines, from optics to accelerometry. In contrast, assessing energy intake still heavily relies on self-report mechanisms. Although these tools have improved, moving from paper-based to online reporting, considerable room for refinement remains in existing tools, and great opportunities exist for novel, transformational tools, including those using spectroscopy and chemo-sensing. This report reviews the state of the science, and the opportunities and challenges in existing and emerging technologies, from the perspectives of 3 key stakeholders: researchers, users, and developers. Each stakeholder approaches these tools with unique requirements: researchers are concerned with validity, accuracy, data detail and abundance, and ethical use; users with ease of use and privacy; and developers with high adherence and utilization, intellectual property, licensing rights, and monetization. Cross-cutting concerns include frequent updating and integration of the food and nutrient databases on which assessments rely, improving accessibility and reducing disparities in use, and maintaining reliable technical assistance. These contextual challenges are discussed in terms of opportunities and further steps in the direction of personalized health.

Reallocation of time between device-measured movement behaviours and risk of incident cardiovascular disease

OBJECTIVE

To improve classification of movement behaviours in free-living accelerometer data using machine-learning methods, and to investigate the association between machine-learned movement behaviours and risk of incident cardiovascular disease (CVD) in adults.

METHODS

Using free-living data from 152 participants, we developed a machine-learning model to classify movement behaviours (moderate-to-vigorous physical activity behaviours (MVPA), light physical activity behaviours, sedentary behaviour, sleep) in wrist-worn accelerometer data. Participants in UK Biobank, a prospective cohort, were asked to wear an accelerometer for 7 days, and we applied our machine-learning model to classify their movement behaviours. Using compositional data analysis Cox regression, we investigated how reallocating time between movement behaviours was associated with CVD incidence.

RESULTS

In leave-one-participant-out analysis, our machine-learning method classified free-living movement behaviours with mean accuracy 88% (95% CI 87% to 89%) and Cohen's kappa 0.80 (95% CI 0.79 to 0.82). Among 87 498 UK Biobank participants, there were 4105 incident CVD events. Reallocating time from any behaviour to MVPA, or reallocating time from sedentary behaviour to any behaviour, was associated with lower CVD risk. For an average individual, reallocating 20 min/day to MVPA from all other behaviours proportionally was associated with 9% (95% CI 7% to 10%) lower risk, while reallocating 1 hour/day to sedentary behaviour from all other behaviours proportionally was associated with 5% (95% CI 3% to 7%) higher risk.

CONCLUSION

Machine-learning methods classified movement behaviours accurately in free-living accelerometer data. Reallocating time from other behaviours to MVPA, and from sedentary behaviour to other behaviours, was associated with lower risk of incident CVD, and should be promoted by interventions and guidelines.

Active Play Network Influences on Physical Activity Among Children Living in Texas Colonias

Physical activity (PA) is beneficial for child health; however, few children meet PA guidelines. Social relationships impact child PA behaviors; however, little is known about this effect in Mexican-heritage populations. This study aims to understand associations between self-reported play networks and PA among Mexican-heritage children. Mexican-heritage children from colonias along the Texas-Mexico border (n = 44; 54.5% girls; mean age = 9.89 years, SD = 0.97) reported information on up to 5 people they played with most often. Linear regression was used to analyze the relationship between composition of children's social network and minutes of moderate- to vigorous-intensity PA (MVPA) and sedentary minutes per day measured by accelerometers. Children who reported a higher percentage of friends as opposed to family members attained significantly more minutes of MVPA per day (β = .27, P = .04). Children who reported playing with individuals in their network more often (β = ‒.28, P = .03) were sedentary for fewer minutes per day. Increasing the connections between children in the neighborhood or community, as well as increasing a child's frequency of active play, may be promising approaches to increasing MVPA and decreasing sedentary behaviors among Mexican-heritage children.

How Accurate and Precise Can We Measure the Posture and the Energy Expenditure Component of Sedentary Behaviour with One Sensor?

Sedentary behaviour is an emergent public health topic, but there is still no method to simultaneously measure both components of sedentary behaviour-posture and energy expenditure-with one sensor. This study investigated the accuracy and precision of measuring sedentary time when combining the proprietary processing of a posture sensor (activPAL) with a new energy expenditure algorithm and the proprietary processing of a movement sensor (ActiGraph) with a published posture algorithm. One hundred office workers wore both sensors for an average of 7 days. The activPAL algorithm development used 38 and the subsequent independent method comparison 62 participants. The single sensor sedentary estimates were compared with Bland-Atman statistics to the Posture and Physical Activity Index, a combined measurement with both sensors. All single-sensor methods overestimated sedentary time. However, adding the algorithms reduced the overestimation from 129 to 21 (activPAL) and from 84 to 7 min a day (ActiGraph), with far narrower 95% limits of agreements. Thus, combining the proprietary data with the algorithms is an easy way to increase the accuracy and precision of the single sensor sedentary estimates and leads to sedentary estimates that are more precise at the individual level than those of the proprietary processing are at the group level.

Wearable energy efficient fitness tracker for sports person health monitoring application

Continuous evaluation of biological and physiological metrics of sports personalities, evaluating general health status, and alerting for life-saving treatments, is supposed to enhance efficiency and healthy performance. Wearable devices with acceptable form factors compact, flexibility, minimal power consumption, etc., are needed for continuous monitoring to avoid affecting everyday operations, thereby retaining functional effectiveness and consumer satisfaction. This research focuses on the acceleration tracker for particularizing the work. Acceleration data is typically collected on battery-powered sensors for activity detection, referring to an exchange between high-precision detection and energy-efficient processing. From a feature selection perspective, the paper explores this trade-off. It suggests an Energy-Efficient Behavior Recognition System with a comprehensive energy utilization model and the Multi-objective Algorithm of Particle Swarm Optimization (EEBRS-MPSO). Therefore, using Random Forest (RF) classifiers, the model and algorithm are tested to measure the precision of identification and obtain the task’s best performance with the lowest energy consumption, among other biologically-inspired algorithms. The findings indicate that energy consumption for data storage and data processing is minimized with magnitude relative to the raw data method by choosing suitable groups of attributes. Thus, the platform allows a scalable range of feature clusters that require the authors to provide an adequate power adjustment for given target use.

GRANADA consensus on analytical approaches to assess associations with accelerometer-determined physical behaviours (physical activity, sedentary behaviour and sleep) in epidemiological studies

The inter-relationship between physical activity, sedentary behaviour and sleep (collectively defined as physical behaviours) is of interest to researchers from different fields. Each of these physical behaviours has been investigated in epidemiological studies, yet their codependency and interactions need to be further explored and accounted for in data analysis. Modern accelerometers capture continuous movement through the day, which presents the challenge of how to best use the richness of these data. In recent years, analytical approaches first applied in other scientific fields have been applied to physical behaviour epidemiology (eg, isotemporal substitution models, compositional data analysis, multivariate pattern analysis, functional data analysis and machine learning). A comprehensive description, discussion, and consensus on the strengths and limitations of these analytical approaches will help researchers decide which approach to use in different situations. In this context, a scientific workshop and meeting were held in Granada to discuss: (1) analytical approaches currently used in the scientific literature on physical behaviour, highlighting strengths and limitations, providing practical recommendations on their use and including a decision tree for assisting researchers’ decision-making; and (2) current gaps and future research directions around the analysis and use of accelerometer data. Advances in analytical approaches to accelerometer-determined physical behaviours in epidemiological studies are expected to influence the interpretation of current and future evidence, and ultimately impact on future physical behaviour guidelines.

Role of AI physical education based on application of functional sports training

The physical health test of college students is an important part of the school physical education work and an important part of the school education evaluation system. It is an educational method that promotes the healthy development of students’ physical fitness and encourages students to actively take physical exercises. It is an individual evaluation standard for students’ physical fitness. It is also one of the necessary conditions for students to graduate. In order to improve the physique and health of college students, this article first introduces functional exercise tests to comprehensively measure the exercise capacity of the main muscle groups and joints of the human body, and integrate flexibility and strength qualities. Secondly, this article quantitatively studies the interaction law between the natural light environment comfort of sports training facilities and architectural design elements, and adopts appropriate dynamic optimization methods to improve the light environment quality of the sports space, thereby enhancing the visual comfort of the sports crowd in the stadium. Finally, the artificial intelligence technology is introduced, through the design of artificial intelligence system, intelligent data collection, and analysis. From the perspective of physical education, the functional exercise test based on artificial intelligence conforms to the essential meaning of the physical fitness test and helps to enhance the awareness of college students’ physical exercise. And the intelligent remote multimedia physical education system based on artificial intelligence makes the physical education process flexible, free from time and place restrictions, and can adopt different teaching strategies according to the different situations of students to implement personalized teaching.

Can an inertial measurement unit (IMU) in combination with machine learning measure fast bowling speed and perceived intensity in cricket?

This study examined whether an inertial measurement unit (IMU), in combination with machine learning, could accurately predict two indirect measures of bowling intensity through ball release speed (BRS) and perceived intensity zone (PIZ). One IMU was attached to the thoracic back of 44 fast bowlers. Each participant bowled 36 deliveries at two different PIZ zones (Zone 1 = 24 deliveries at 70% to 85% of maximum perceived bowling effort; Zone 2 = 12 deliveries at 100% of maximum perceived bowling effort) in a random order. IMU data (sampling rate = 250 Hz) were downsampled to 125 Hz, 50 Hz, and 25 Hz to determine if model accuracy was affected by the sampling frequency. Data were analysed using four machine learning models. A two-way repeated-measures ANOVA was used to compare the mean absolute error (MAE) and accuracy scores (separately) across the four models and four sampling frequencies. Gradient boosting models were shown to be the most consistent at measuring BRS (MAE = 3.61 km/h) and PIZ (F-score = 88%) across all sampling frequencies. This method could be used to measure BRS and PIZ which may contribute to a better understanding of overall bowling load which may help to reduce injuries.

Moving the Lab into the Mountains: A Pilot Study of Human Activity Recognition in Unstructured Environments

Goal: To develop and validate a field-based data collection and assessment method for human activity recognition in the mountains with variations in terrain and fatigue using a single accelerometer and a deep learning model. Methods: The protocol generated an unsupervised labelled dataset of various long-term field-based activities including run, walk, stand, lay and obstacle climb. Activity was voluntary so transitions could not be determined a priori. Terrain variations included slope, crossing rivers, obstacles and surfaces including road, gravel, clay, mud, long grass and rough track. Fatigue levels were modulated between rested to physical exhaustion. The dataset was used to train a deep learning convolutional neural network (CNN) capable of being deployed on battery powered devices. The human activity recognition results were compared to a lab-based dataset with 1,098,204 samples and six features, uniform smooth surfaces, non-fatigued supervised participants and activity labelling defined by the protocol. Results: The trail run dataset had 3,829,759 samples with five features. The repetitive activities and single instance activities required hyper parameter tuning to reach an overall accuracy 0.978 with a minimum class precision for the one-off activity (climbing gate) of 0.802. Conclusion: The experimental results showed that the CNN deep learning model performed well with terrain and fatigue variations compared to the lab equivalents (accuracy 97.8% vs. 97.7% for trail vs. lab). Significance: To the authors knowledge this study demonstrated the first successful human activity recognition (HAR) in a mountain environment. A robust and repeatable protocol was developed to generate a validated trail running dataset when there were no observers present and activity types changed on a voluntary basis across variations in terrain surface and both cognitive and physical fatigue levels.

Adoption of Machine Learning Algorithm-Based Intelligent Basketball Training Robot in Athlete Injury Prevention

In order to effectively prevent sports injuries caused by collisions in basketball training, realize efficient shooting, and reduce collisions, the machine learning algorithm was applied to intelligent robot for path planning in this study. First of all, combined with the basketball motion trajectory model, the sport recognition in basketball training was analyzed. Second, the mathematical model of the basketball motion trajectory of the shooting motion was established, and the factors affecting the shooting were analyzed. Thirdly, on this basis, the machine learning-based improved Q-Learning algorithm was proposed, the path planning of the moving robot was realized, and the obstacle avoidance behavior was accomplished effectively. In the path planning, the principle of fuzzy controller was applied, and the obstacle ultrasonic signals acquired around the robot were taken as input to effectively avoid obstacles. Finally, the robot was able to approach the target point while avoiding obstacles. The results of simulation experiment show that the obstacle avoidance path obtained by the improved Q-Learning algorithm is flatter, indicating that the algorithm is more suitable for the obstacle avoidance of the robot. Besides, it only takes about 250 s for the robot to find the obstacle avoidance path to the target state for the first time, which is far lower than the 700 s of the previous original algorithm. As a result, the fuzzy controller applied to the basketball robot can effectively avoid the obstacles in the robot movement process, and the motion trajectory curve obtained is relatively smooth. Therefore, the proposed machine learning algorithm has favorable obstacle avoidance effect when it is applied to path planning in basketball training, and can effectively prevent sports injuries in basketball activities.

Calibrating Wrist-Worn Accelerometers for Physical Activity Assessment in Preschoolers: Machine Learning Approaches

BACKGROUND

Physical activity (PA) level is associated with multiple health benefits during early childhood. However, inconsistency in the methods for quantification of PA levels among preschoolers remains a problem.

OBJECTIVE

This study aimed to develop PA intensity cut points for wrist-worn accelerometers by using machine learning (ML) approaches to assess PA in preschoolers.

METHODS

Wrist- and hip-derived acceleration data were collected simultaneously from 34 preschoolers on 3 consecutive preschool days. Two supervised ML models, receiver operating characteristic curve (ROC) and ordinal logistic regression (OLR), and one unsupervised ML model, k-means cluster analysis, were applied to establish wrist-worn accelerometer vector magnitude (VM) cut points to classify accelerometer counts into sedentary behavior, light PA (LPA), moderate PA (MPA), and vigorous PA (VPA). Physical activity intensity levels identified by hip-worn accelerometer VM cut points were used as reference to train the supervised ML models. Vector magnitude counts were classified by intensity based on three newly established wrist methods and the hip reference to examine classification accuracy. Daily estimates of PA were compared to the hip-reference criterion.

RESULTS

In total, 3600 epochs with matched hip- and wrist-worn accelerometer VM counts were analyzed. All ML approaches performed differently on developing PA intensity cut points for wrist-worn accelerometers. Among the three ML models, k-means cluster analysis derived the following cut points: ≤2556 counts per minute (cpm) for sedentary behavior, 2557-7064 cpm for LPA, 7065-14532 cpm for MPA, and ≥14533 cpm for VPA; in addition, k-means cluster analysis had the highest classification accuracy, with more than 70% of the total epochs being classified into the correct PA categories, as examined by the hip reference. Additionally, k-means cut points exhibited the most accurate estimates on sedentary behavior, LPA, and VPA as the hip reference. None of the three wrist methods were able to accurately assess MPA.

CONCLUSIONS

This study demonstrates the potential of ML approaches in establishing cut points for wrist-worn accelerometers to assess PA in preschoolers. However, the findings from this study warrant additional validation studies.

Wristbands Containing Accelerometers for Objective Arm Swing Analysis in Patients with Parkinson's Disease

In patients with Parkinson’s disease (PD), arm swing changes are common, even in the early stages, and these changes are usually evaluated subjectively by an expert. In this article, hypothesize that arm swing changes can be detected using a low-cost, cloud-based, wearable, sensor system that incorporates triaxial accelerometers. The aim of this work is to develop a low-cost, assistive diagnostic tool for use in quantifying the arm swing kinematics of patients with PD. Ten patients with PD and 11 age-matched, healthy subjects are included in the study. Four feature extraction techniques were applied: (i) Asymmetry estimation based on root mean square (RMS) differences between arm movements; (ii) posterior–anterior phase and cycle regularity through autocorrelation; (iii) tremor energy, established using Fourier transform analysis; and (iv) signal complexity through the fractal dimension by wavelet analysis. The PD group showed significant (p < 0.05) reductions in arm swing RMS values, higher arm swing asymmetry, higher anterior–posterior phase regularities, greater “high energy frequency” signals, and higher complexity in their XZ plane signals. Therefore, the novel, portable system provides a reliable means to support clinical practice in PD assessment.

Development of cut-points for determining activity intensity from a wrist-worn ActiGraph accelerometer in free-living adults

Despite recent popularity of wrist-worn accelerometers for assessing free-living physical behaviours, there is a lack of user-friendly methods to characterize physical activity from a wrist-worn ActiGraph accelerometer. Participants in this study completed a laboratory protocol and/or 3-8 hours of directly observed free-living (criterion measure of activity intensity) while wearing ActiGraph GT9X Link accelerometers on the right hip and non-dominant wrist. All laboratory data (n = 36) and 11 participants' free-living data were used to develop vector magnitude count cut-points (counts/min) for activity intensity for the wrist-worn accelerometer, and 12 participants' free-living data were used to cross-validate cut-point accuracy. The cut-points were: <2,860 counts/min (sedentary); 2,860-3,940 counts/min (light); and ≥3,941counts/min (moderate-to-vigorous (MVPA)). These cut-points had an accuracy of 70.8% for assessing free-living activity intensity, whereas Sasaki/Freedson cut-points for the hip accelerometer had an accuracy of 77.1%, and Hildebrand Euclidean Norm Minus One (ENMO) cut-points for the wrist accelerometer had an accuracy of 75.2%. While accuracy was higher for a hip-worn accelerometer and for ENMO wrist cut-points, the high wear compliance of wrist accelerometers shown in past work and the ease of use of count-based analysis methods may justify use of these developed cut-points until more accurate, equally usable methods can be developed.