Wearable preimpact fall detector using SVM

In order to distinguish falls from normal activities exactly, a fall early warning wearable detector combining angle with acceleration features was proposed in this paper. The detector consists of MEMS inertial sensor and smartphone. The application to solve classification algorithm using Support Vector Machine is developed. Experimental trials which young adults participated in involved 250 falls (4 types, forward, backward, left and right) and 250 normal activities (10 types, bowing, jogging, ascending stairs, etc.). The results of experiment showed the detector provided a sensitivity of 99%, a specificity of 96.5% and the average lead-time is 268 ms. The approached detector's feasibility and efficiency in detecting falls from daily events were verified.