Fall Detection Algorithm based on Gradient Boosting Decision Tree
Yunkun Ning, Sheng Zhang, Xiaofen Nie, Guanglin Li, Guoru Zhao · 2019
This paper proposes a fall detection algorithm for wearable devices or mobile terminals. Since there is currently no complete data set for the fall of the elderly, most of the research uses young experimenters to collect data, so the data set and the difference in the terminal leads to the low accuracy of the fall detection algorithm. In order to improve the accuracy of the algorithm, this paper uses a two-level detection that combines the offline threshold method with the online GBDT (Gradient Boosting Decision Tree) to detect the fall action. According to the analysis of the final result, the sensitivity of the GBDT algorithm is 96%, the false positive rate is 2.8%, and the accuracy rate is 96%. The improved decision tree algorithm can detect the fall behavior more accurately than the decision tree.