Active learning with extreme learning machine for online imbalanced multiclass classification

Characterization of noise in long-term ECG monitoring with machine learning based on clinical criteria

Noise and artifacts affect strongly the quality of the electrocardiogram (ECG) in long-term ECG monitoring (LTM), making some of its parts impractical for diagnosis. The clinical severity of noise defines a qualitative quality score according to the manner clinicians make the interpretation of the ECG, in contrast to assess noise from a quantitative standpoint. So clinical noise refers to a scale of different levels of qualitative severity of noise which aims at elucidating which ECG fragments are valid to achieve diagnosis from a clinical point of view, unlike the traditional approach, which assesses noise in terms of quantitative severity. This work proposes the use of machine learning (ML) techniques to categorize different qualitative noise severity using a database annotated according to a clinical noise taxonomy as gold standard. A comparative study is carried out using five representative ML methods, namely, K neareast neighbors, decision trees, support vector machine, single-layer perceptron, and random forest. The models are fed by signal quality indexes characterizing the waveform in time and frequency domains, as well as from a statistical viewpoint, to distinguish between clinically valid ECG segments from invalid ones. A solid methodology to prevent overfitting to both the dataset and the patient is developed, taking into account balance of classes, patient separation, and patient rotation in the test set. All the proposed learning systems have demonstrated good classification performance, attaining a recall, precision, and F1 score up to 0.78, 0.80, and 0.77, respectively, in the test set by a single-layer perceptron approach. These systems provide a classification solution for assessing the clinical quality of the ECG taken from LTM recordings. Graphical Abstract Clinical Noise Severity Classification based on Machine Learning techniques towards Long-Term ECG Monitoring.

Improving Active Learning Performance through the Use of Data Augmentation

Active learning (AL) is a well-known technique to optimize data usage in training, through the interactive selection of unlabeled observations, out of a large pool of unlabeled data, to be labeled by a supervisor. Its focus is to find the unlabeled observations that, once labeled, will maximize the informativeness of the training dataset, therefore reducing data-related costs. The literature describes several methods to improve the effectiveness of this process. Nonetheless, there is a paucity of research developed around the application of artificial data sources in AL, especially outside image classification or NLP. This paper proposes a new AL framework, which relies on the effective use of artificial data. It may be used with any classifier, generation mechanism, and data type and can be integrated with multiple other state-of-the-art AL contributions. This combination is expected to increase the ML classifier’s performance and reduce both the supervisor’s involvement and the amount of required labeled data at the expense of a marginal increase in computational time. The proposed method introduces a hyperparameter optimization component to improve the generation of artificial instances during the AL process as well as an uncertainty-based data generation mechanism. We compare the proposed method to the standard framework and an oversampling-based active learning method for more informed data generation in an AL context. The models’ performance was tested using four different classifiers, two AL-specific performance metrics, and three classification performance metrics over 15 different datasets. We demonstrated that the proposed framework, using data augmentation, significantly improved the performance of AL, both in terms of classification performance and data selection efficiency (all the codes and preprocessed data developed for this study are available at https://github.com/joaopfonseca/publications/).

Fault diagnosis and prediction of wind turbine gearbox based on a new hybrid model

Gearbox is an important part of wind turbine. Diagnosing and forecasting gearbox faults of wind turbines can effectively reduce the costs of operation and maintenance and improve the reliability of gearbox operation. Due to high dimensionality and nonlinearity of system parameters, the paper uses the grey relation analysis to select features related to gearbox oil temperature. Features with a relational degree above 0.7 are selected as input data related to oil temperature, including wind speed, ambient temperature, power, and gearbox shaft temperature. Then, a new extreme learning machine with kernel improved by the whale optimization algorithm is established to forecast gearbox oil temperature. Through the residuals between gearbox oil temperature predicted by the proposed model and monitored by the SCADA, whether the gearbox exists faults can be diagnosed. In the case study, the test data was divided into two groups (the test data with and without faults). In the data test without faults, compared with three other models, the proposed model has the smallest false-negative rate (0.211%) and mean absolute percentage error (2.812%). In the data test with faults, the proposed model can diagnose gearbox faults earlier (160 min in advance) than the other three benchmark models. The results show that the proposed hybrid model performs well in the fault diagnosis and prediction of wind turbine gearbox.

Methodology for Interactive Labeling of Patched Asphalt Pavement Images Based on U-Net Convolutional Neural Network

Image recognition based on deep learning generally demands a huge sample size for training, for which the image labeling becomes inevitably laborious and time-consuming. In the case of evaluating the pavement quality condition, many pavement distress patching images would need manual screening and labeling, meanwhile the subjectivity of the labeling personnel would greatly affect the accuracy of image labeling. In this study, in order for an accurate and efficient recognition of the pavement patching images, an interactive labeling method is proposed based on the U-Net convolutional neural network, using active learning combined with reverse and correction labeling. According to the calculation results in this paper, the sample size required by the interactive labeling is about half of the traditional labeling method for the same recognition precision. Meanwhile, the accuracy of interactive labeling method based on the mean intersection over union (mean_IOU) index is 6% higher than that of the traditional method using the same sample size and training epochs. In addition, the accuracy analysis of the noise and boundary of the prediction results shows that this method eliminates 92% of the noise in the predictions (the proportion of noise is reduced from 13.85% to 1.06%), and the image definition is improved by 14.1% in terms of the boundary gray area ratio. The interactive labeling is considered as a significantly valuable approach, as it reduces the sample size in each epoch of active learning, greatly alleviates the demand for manpower, and improves learning efficiency and accuracy.