Analysis of a Feature Incremental Learning Method for Sensor-based Activity Recognition

The recognition of day-to-day activities is a major research subject for the monitoring of health and elderly care. Most existing models, though, operate only in static and predefined sensor configurations. It is a significant challenge to adapt an existing activity identification model to the development of new sensors in a dynamic environment to monitor the Health. In this paper, its proposed to increase the performance of an existing model with small figures on newly developed elements, a new incremental method of learning that includes the Incremental Random Forest Function, two important elements, 1) Mutual Diversity Knowledge Strategy (MIDGS) and 2) Incremental Tree Cultivation Mechanism, feature incremental tree growing mechanism (FITGM). MIDGS improves the internal random forest diversity while FITGM improves the exactness of individual decision-making areas. To assess feature incremental random forest (FIRF) performance, extensive experiments performance done for activity recognition on three well-known public data set. Experimental findings suggest that FIRF is much more reliable and effective than the other state-of-the-art methods. This allows new sensors in evolving environments to be controlled dynamically.