Body Sensor System for Health Support based on Machine Learning

Abstract

A system which recognizes the activities during desk work and study using unobtrusive wearable sensors was developed to support health appropriately depending on the individuals' states. The glasses detect head motions as well as eye movements and blinks. The watch detects wrist motions as well as pulse rate. The data are sent to the developed app on a smartphone from the multiple sensors via Bluetooth LE and processed. The effective multi-class classifiers and feature selection methods were examined to classify the activities with high-accuracy. GA-SVM had the highest accuracy to classify four states during study. Similarly, the activities during desk work were effectively classified using RF. Therefore, health support could be given based on the relationship between the activities and body conditions such as posture and eye fatigue utilizing the system.