Assessing Stroke Patients Movements Using Inertial Measurements Through the Advances of Ensemble Learning Technology
Najmeh Razfar, Rasha Kashef, Farah A. Mohammadi · 2021
Internet of things (IoT) and wearable sensors enabled the possibility of measuring activities of daily living (ADLs). The accuracy and precision of detecting stroke patients' movements, using wearable sensors and analyzing data by implementing the advancement of machine learning and ensemble learning technology can enhance the stroke patients' remote assessment techniques. Therefore, this paper aimed to apply ensemble in addition to the machine learning models on the Xsens sensors dataset derived from wearable sensors and collected from twenty stroke survivors. Then, we compare the performance of the single model with the bagging and boosting ensemble learning methods to detect the affected hand of the stroke survivors from the non-affected hand. The results indicated that the ensemble techniques such as GentleBoost and AdaBoost achieved the highest model performance with 91.2% and 89.6% accuracy, respectively in comparison with other single machine learning techniques as well as other ensembles. It was also noted that the subspace ensemble techniques achieved the lowest accuracy, recall, and specificity compared to the Decision Tree.