A Machine Learning Approach to Human Activity Recognition

Abstract

Human Activity Recognition (HAR) is the problem of classifying an individual's activity into well-defined moments, utilizing responsive sensors that are influenced by human movement. Sensor-enabled smartphones make Human Activity Recognition progressively significant and well known. The physical sensors, gyroscope and accelerometer combinedly allow the devices to provide motion measuring capabilities in a more accurate manner. The present research work adopts a machine learning based approach for recognizing activity on the basis of data collected through the smartphone sensors (accelerometer and gyroscope). Various state-of-the-art machine learning based techniques have been employed and compared on the basis of the performance metrics, accuracy, recall, precision, and the F1-score. Of all the selected different machine learning classifiers, the best result is given by the Support Vector Machine (SVM) with ‘RBF’ kernel, which achieved an accuracy of 96.61 % in classifying the activities into the six different classes.