Feature Selection and Classification on Human Activity Recognition using Machine Learning Approaches

Human Activity Recognition (HAR) is a well-researched area that aims to recognize the activity performed by a person. But the practical applications often encounter complications such as “The curse of dimensionality� and “Redundant features� which results in their poor performance. Hence, the need for feature selection is very imperative in such cases. This paper aims at identifying the subsets of HAR dataset that consists of most important and relevant features using Boruta Feature Selection Algorithm. Because the dataset with less number of features that are more relevant requires less computational time to train the classifier and it also improves the accuracy rate of the classification model. Upon identification, this paper also implements Support Vector Machines (SVMs) classification algorithm on identified subsets as well as HAR dataset. Furthermore, comparing the accuracy rates attained by the classifier on different subsets as well as their computational time.