Deep Learning-Based Human Activity Recognition From Wearable Devices

Abstract

The rise of sedentary lifestyles and the prevalence of lifestyle-related health issues highlight the need for personalized wellness solutions to encourage physical activity and maintain health. Human fitness activities, ranging from basic exercises like jumping jacks and squats to advanced routines such as pull-ups, are key to promoting physical fitness and individualized wellness. Historically, fitness routines were guided by general recommendations or manual tracking methods, which lacked precision and personalization. Traditional systems, such as fitness logs, generic plans, and fitness assessments, offered limited insights, were labor-intensive, error-prone, and lacked adaptability. This research leverages artificial intelligence (AI) and machine learning (ML) to overcome the limitations of traditional systems. Advances in wearable technology, sensor data, and large-scale datasets enable intelligent systems for multi-class classification of human fitness activities, providing personalized feedback, performance evaluation, and adaptive fitness plans tailored to individual needs. Our current system uses K-Nearest Neighbors (KNN) and LightGBM (LGBM) to classify fitness activities; however, these models face limitations in handling complex patterns and large-scale data. To address this, we propose a shift to Deep Neural Networks (DNNs), which better capture intricate patterns in high-dimensional data, offering higher accuracy, scalability, and adaptability for personalized wellness. By integrating DNNs, the system will enhance activity recognition, providing more accurate insights and highly personalized recommendations, offering a robust, scalable solution for improving fitness routines and fostering healthier lifestyles, and redefining fitness monitoring to enable long-term wellness.