A Machine Learning Framework for Edge Computing to Improve Prediction Accuracy in Mobile Health Monitoring

The great challenges in the aging society and the lack of human resources, especially in health care, remains a formidable task. The cloud centric computing paradigm offers a solution in processing IoT applications in health care. However, due to the large computing and communication overheads, an alternative solution is sought. Here, we consider machine learning in edge computing to detect and improve the predictability accuracy in mobile health monitoring of human activity. With multi-modal sensor data, we conducted preprocessing to sanitize the data and classify the activities in the dataset. We used and compare the processing performance using random forests and SVM machine learning algorithms to identify and classify the activities in the dataset. We achieved approximately 99% accuracy with random forest which was better than SVM, at 98%. We used confusion matrix to identify the majority of mismatched data belonging to initial value of sensors while recording a particular activity, and also used visual representation of the data for better understanding. We extract the activity's ECG data and classify into four categories to provide more specific information from the person's activity data. The aforementioned experiments provided promising results and insights on the implementation to improve the prediction accuracy on the health status of people undergoing some activity.