An improved algorithm for human activity recognition using wearable sensors
Ye Chen, Ming Guo, Zhelong Wang · 2016
In this paper, a novel approach is investigated to recognize human activities by using wearable sensors. Three key techniques are mainly discussed including the ensemble empirical mode decomposition (EEMD), the sparse multinomial logistic regression algorithm with Bayesian regularization (SBMLR) and the fuzzy least squares support vector machine (FLS-SVM). All of the features based on the EEMD are extracted from sensor data. Then, the features vectors are processed by an embedded feature selection algorithm - SBMLR, which may remarkably reduce the dimension and maintain the most discriminative information. The FLS-SVM technique is employed to deal with the reduced features and identify human activities. Experimental results show that our approach achieves an overall mean classification rate of 93.43%, which exhibits the remarkable recognition performance compared with other approaches. We conclude that the proposed approach could play an important role in human activity recognition (HAR) using wearable sensors, especially in real-time applications and large-scale dataset processing.