A Sensor-Based Human Activity Recognition System via Restricted Boltzmann Machine and Extended Space Forest

Abstract

This paper presents a classification system for activity recognition (AR) based on information gained from multi-sensors. Normally, the activity data received from different sensors are employed to construct features with high dimensionality. To automatically extract informative features from complex activities data set, an approach integrating feature extraction and ensemble learning is designed. Specifically, the restricted Boltzmann machines (RBM) and extended space forest (ESF) algorithms are combined in a suitable manners to generate accurate and diverse classifiers. The system conducts experiments on two real-world activity recognition data sets and the results show the effectiveness of the proposed system.