A low-cost fall detection system using SVM algorithm based on G-Fall dataset collected by wearable device
Yuan Chen, Zhaoliang Guan, Hangyu Xu, Jinkun Ke, Ping Chen, Chun Zhang · 2023
Fall detection using wearable devices has the advantage of high accuracy and low cost. This paper discloses a large G-sensor dataset that relies on a self-developed wearable device that collects acceleration data from the waist, thigh, and wrist during daily activities and when falls occur. In this dataset, a total of 24 volunteers participated, including 15 types of daily actions and 11 types of fall actions. After data preprocessing, we experimented with four methods: threshold method, support vector machine (SVM), K-nearest neighbor (KNN), and artificial neural network (ANN), and concluded that the waist was the best collection location and SVM algorithm was the optimal method. When using the support vector machine (SVM) for fall detection at the waist, the average accuracy of classifying daily actions and fall actions can reach 98.61%. And when using it to classify 26 types of actions, the average accuracy can reach 90.03%.