A Federated Learning Based Hybrid Deep Learning Framework for Enhanced Human Activity Recognition

Human Activity Recognition (HAR) has become increasingly important with the advent of mobile computing and sensor technologies. Traditional HAR systems, relying on centralized data processing and supervised learning techniques, face significant challenges related to data privacy, scalability, and the need for extensive labeled datasets. Federated Learning (FL) has emerged as a promising solution to address these limitations by allowing collective model training without centralizing sensitive user data, thereby enhancing privacy and personalization. This paper introduces a novel framework that integrates FL with hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) models to improve the accuracy and robustness of HAR. Our approach leverages the strengths of CNNs in capturing spatial features and LSTMs in modeling temporal dependencies, while maintaining data privacy through FL. Extensive experiments on real-world HAR datasets demonstrate that our proposed framework not only preserves privacy but also achieves high recognition rates with limited labeled data, showcasing its potential for practical applications in healthcare, fitness, and smart home environments.