Time-Distributed Layer Convolutions with Long Short-Term Memory for Human Activity Recognition and Result Comparison with Various Machine Learning Models
Jaykumar S. Dhage, Avinash K. Gulve · Engineering Technology & Applied Science Research · 2025
Human Activity Recognition (HAR) is an essential area of research with many applications in healthcare, security, and entertainment. One of the main challenges in HAR is the variability in human behavior and reactions to similar inputs, which complicates accurate prediction. This study investigates the utilization of deep learning techniques in enhancing the HAR accuracy. The proposed method uses a Time-Distributed Layer (TDL) framework with LSTM to achieve effective feature extraction and temporal pattern recognition from sensor data. The proposed approach was compared with traditional machine learning models, such as Logistic Regression (LR), Support Vector Machines (SVM), Decision Trees (DT), and Random Forests (RF), to evaluate its effectiveness. The experimental results demonstrate that deep learning models significantly outperform traditional approaches, achieving 97.57% accuracy with TDL-LSTM and 97.81% accuracy with LSTM-TDL, while conventional methods exhibit lower performance. The comparison highlights the advantages of deep learning methods in capturing both spatial and temporal dependencies, resulting in more robust HAR systems. Overall, this study demonstrates the superiority of LSTM-based architectures over traditional models, paving the way for future advances in real-world HAR applications, including wearable devices and intelligent monitoring systems.