Classification of Human Activity Recognition Using Machine Learning on the WISDM Dataset

Abstract

The significance of human activity recognition (HAR) is rising as it seeks to improve everyday life and healthcare through better technology access and efficiency. Its objective is to transform industries by enabling smart homes, improving robots, bolstering security, and improving human-computer interactions. HAR works to improve well-being, which is essential to health, wellness, and sports. While the complexity of human behavior poses challenges, machine learning advancements offer hope for solutions. Continuous research in accurately detecting a wide array of human activities underscores the significant impact of HAR on technological development and its broad applications. In this work, a convolution neural network CNN algorithm and random forest RF algorithms were produced for human recognition activity classification using WISDM-51 dataset that contains 18 human activities. The CNN achieved an accuracy of 89.36%, whereas the RF algorithm reached a slightly higher accuracy of 93.46%. The results suggest that the proposed algorithms offer promising potential.

Keywords

• HAR, CNN, RF, machine learning