An Enhanced Fall Detection Approach Based on Cost Sensitivity Analysis

Falling may cause serious injury, especially for the elderly, it had been proven that fall detection system using wearable devices based on tri-axial accelerator is effective and feasible. Most of the time, fall do not lead damage immediately, but the psychological burden after fall and other environment factors, could bring on serious damage when the elderly cannot get timely help. To deal with these problems, we propose an enhanced approach for fall detection based on the time series data generated by the tri-axial accelerator. Two approaches of cost sensitivity analysis are presented, including minimal risk Bayes and Neyman-Pearson method, both of them lead to a significant sensitivity enhancement. In other words, the proposed strategies mean a significant reduce in collectivity decision making risk stage, and the true non-fall activities recognition rate will be improved by via validation after classification.