Wearable Sensors based Exertion Recognition using Statistical Features and Random Forest for Physical Healthcare Monitoring

Remote physical activity recognition is gaining popularity as it provides improved healthcare monitoring services without hampering the daily lifestyle of individuals. For smart healthcare, several wearable sensors i.e., inertial measurement unit (IMU), mechanomyography (MMG), electromyography (EMG) and other biosignal devices are used commonly to improve quality of life. In this paper, we have proposed an efficient model including multiple domain features, feature reduction and recognizer engine to provide improved healthcare monitoring. Time-domain and statistical features have been integrated with the system to ensure robustness as a performance measure. These features include zero crossings, abrupt changes, peak to peak, and crest factor. For reducing the dimensionality, feature selection methods i.e. minimal-redundancy-maximal-relevance (MRMR) and novel multi-layer sequential forward selection (MLSFS) have been considered with symbol selection. Improved classification results have been achieved by using bagged random forest technique. This model provides unique prospect to researchers by proposing monitoring mechanism for health disorder i.e., asthenia.