ATTENTION-ENHANCED CNN-LSTM MODELS FOR SENSOR-BASED HUMAN ACTIVITY RECOGNITION: A COMPARATIVE STUDY
International Journal of Mechatronics and Applied Mechanics · 2025
Human Activity Recognition (HAR) using wearable sensors is an essential component in healthcare, rehabilitation, and assistive robotics.It enables intelligent systems to interpret and respond to human motion in real time.However, designing models that effectively capture both spatial and temporal dependencies in sensor data remains a major challenge, particularly when aiming for deployment in practical, real-world scenarios.This paper presents a comprehensive comparative study of various deep learning architectures for sensor-based HAR, including Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, and hybrid CNN-LSTM models with and without integrated attention mechanisms.The motivation behind this paper stems from the limitations of conventional machine learning approaches, which require manual feature engineering and struggle with modeling complex, sequential activity transitions.To overcome these challenges, we evaluate each model under consistent conditions using the ENABL3S dataset, a publicly available benchmark that captures bilateral neuromechanical signals from wearable sensors across multiple locomotor tasks.All models were trained using a standardized pipeline, including fixed window segmentation, identical hyperparameters, and equal epoch counts.The comparative results reveal that the TimeDistributed CNN-LSTM model enhanced with a Keras-based attention layer-applied after the LSTM blockachieved the highest test accuracy of 98.91%.The attention mechanism enabled the network to prioritize critical temporal segments in the sensor data, leading to improved classification accuracy.These findings highlight the importance of architectural enhancements in HAR systems and suggest that attention-augmented hybrid models are particularly well-suited for applications demanding high accuracy, such as intelligent prosthetics, real-time rehabilitation monitoring, and personalized mobility support technologies.