Analysis and Effect of Feature Selection Over Smartphone-Based Dataset for Human Activity Recognition

Abstract

The availability of diverse and powerful sensors that are embedded in modern smartphones has created exciting opportunities for developing context-aware services and applications. For example, Human activity recognition (HAR) is an important feature that could be applied to many applications and services, such as those in healthcare and transportation. However, recognizing relevant human activities using smartphones remains a challenging task and requires efficient data mining approaches. In this paper, we present a comparison study for HAR using features selection methods to reduce the training and classification time while maintaining significant performance. In fact, due to the limited resources of Smartphones, reducing the feature set helps reducing computation costs, especially for real-time continuous online applications. We validated our approach on a publicly available dataset to classify six different activities. Results show that Recursive Feature Elimination algorithm works well with Radial Basis Function Support Vector Machine and significantly improves model building time without decreasing recognition performance.