Advanced Human Activity Recognition Using Wearable Sensors

Abstract

Wearable sensor-based Human Activity Recognition (HAR) has emerged as a key area in pervasive computing, healthcare monitoring, and smart environments. With the advancement of low-cost, energy-efficient sensors such as accelerometers and gyroscopes, continuous human motion tracking has become more feasible. Traditional HAR systems relied on manual feature extraction and classical machine learning models like SVM, Decision Trees, and k-NN, but faced challenges such as noise, variability, and computational constraints. This study reviews pre-2018 developments and proposes an improved HAR framework incorporating advanced feature engineering, sensor fusion, and ensemble learning techniques. The system follows key stages including data acquisition, preprocessing, segmentation, feature extraction, and classification. By combining multi-sensor data and hybrid models, the framework enhances classification accuracy and robustness. Evaluation using benchmark datasets like UCI HAR and WISDM demonstrates improved performance over conventional methods. The paper also highlights key challenges such as energy efficiency, scalability, real-time processing, and privacy, while emphasizing the future role of deep learning and adaptive systems for personalized activity recognition. Overall, wearable sensor-based HAR shows strong potential in healthcare, fitness, and smart environments.