Redefining human activity recognition with LSTM and GRU networks
D. Vasumathi, Sreeja S. Sai, P. Anusha, Bindu B. Mounika, M Sirisha · 2025
Recognition of human activity is an important field in artificial intelligence, machine learning, and deep learning (DL), applied in healthcare, sports analysis, security, and human-computer interaction. HAR utilizes sensor data like accelerometers, gyroscopes, cameras, and audio sensors to identify human actions. Deep learning, including convolutional neural networks and recurrent neural networks (RNNs), autonomously learns and classifies activities. Data preprocessing and feature engineering are crucial. Validation ensures real-world accuracy. DL’s continuous evolution promises Human Activity Recognition (HAR) advancements across domains. In our proposed model we used the LSTM (Long Short-Term Memory) and GRU (Gated Recurrent Unit) Algorithm for Human Activity Recognition. The growing dataset volume necessitates machine learning. Our LSTM and GRU, RNN-based model outperform existing methods on HAR datasets. This research paper presents a comprehensive investigation into HAR using Deep learning techniques. We explore data collection methods, feature engineering approaches, and various DL models to accurately classify human activities. Our study includes an in-depth analysis of performance metrics, model evaluation, and real-time applications. Through extensive experiments on diverse datasets, we demonstrate the effectiveness of DL-based HAR systems in achieving high accuracy and potential for real-world deployment. This research contributes to the ongoing efforts in developing robust and versatile HAR solutions.