A Multi-classifier Combination Method Using SFFS Algorithm for Recognition of 19 Human Activities

Abstract

In order to investigate the human activity recognition and classification, which is significant for human-computer interaction (HCI), a multi-classifier combination method using Sequential Forward Feature Selection (SFFS) algorithm is proposed in this paper for recognition of 19 human daily and sports activities. The dataset collected by wearable sensor units is obtained from UCI Machine Learning Repository. The main contents of this method include: (1) extracting features from the raw sensor data after preprocessing; (2) reducing features by SFFS algorithm; (3) classifying activities by 10-fold cross validation with a multi-classifier combination algorithm based on the grid search for parameter optimization. The experimental results indicate that, compared with other traditional activity recognition methods, which use principal component analysis (PCA) to reduce features or use a single classifier to classify activities, the multi-classifier combination method using SFFS achieves the best recognition performance with the average classification accuracy of 99.91 %.