Adaptive multiple classifiers fusion for inertial sensor based human activity recognition

摘要: Aiming at the poor accuracy of single classifier in recognizing daily activities based on accelerometer, this paper presents a method activity recognition ensemble learning and full information matrix fusion weight. Firstly, features from three attributes are extracted acceleration signals respectively. The kinds can well describe activity, they relatively independent, which reduce interference caused by redundancy process fusion. Then base classifiers support vector machines constructed Secondly, Euclidean distance between test sample every training for each type feature is calculated to find out k nearest neighbors set K-nearest neighbour method. cluster analysis used compute similarity neighbor sample. Then, proper threshold utilized remove invalid whose less than threshold. According effective neighbor, calculate accuracy. weight dynamically according Experiments showed that our proposed get best average 94.79% among several other functions when using majority voting method, besides, time cost also appealing.