Human Activity Recognition Using Hybrid CNN-RNN Architecture
Adarsh Muralidharan, Sazia Mahfuz · Procedia Computer Science · 2025
Human Activity Recognition (HAR) is a critical challenge faced in effective healthcare monitoring in nursing homes caring for elderly and vulnerable patients. This research explores the development and evaluation of deep learning models to mitigate the HAR challenges. Given the increasing importance of accurately identifying and classifying human activities for health monitoring, we create a model that integrates Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNNs) in a hybrid architecture to process and analyze sensor data from accelerometers and gyroscopes effectively. Our model aims to achieve high performance for different age groups and physical conditions. Through a comprehensive methodology that includes data preprocessing, feature engineering, and iterative model refinement, we assess the model’s performance against other state-of-the-art models. The final model achieves an F1 score of 96.75%, demonstrating a promising direction toward achieving reliable HAR in real-time applications. This study advances the development of HAR technologies and may significantly benefit patient care by improving health monitoring and intervention capabilities.