Comparative Analysis of Deep Learning and Machine Learning Models for Human Activity Recognition Using Wearable Sensors
R Janaane, S Preethika, Nigade Shweta, A R Nithiyaishwarya, Sivakumar Rajagopal, Rahul Soangra · 2025
Wearable sensors are essential for tracking human activity, but analyzing their data requires robust classification methods. This study evaluates traditional machine learning models (Random Forest, XGBoost) and deep learning models (CNNs, CDAEs) on data from a single sensor across two datasets (30 and 50 individuals). Data preprocessing included feature scaling, alignment, encoding, and SMOTE for class imbalance. Random Forest and XGBoost consistently outperformed other models, achieving 75–77% accuracy, with stable performance across both datasets. Merging datasets improved precision for several models, highlighting the value of larger, diverse data. These findings support the use of ensemble classifiers for reliable, real-time activity recognition in wearable applications.