A Deep Learning-based Hybrid CNN-LSTM Model for Human Activity Recognition

Abstract

Human Activity Recognition (HAR) involves identifying and classifying physical activities performed by individuals using data collected from sensors like accelerometers, gyroscopes, or cameras. HAR has broad applications in healthcare, fitness monitoring, and human-computer interaction, where accurate activity recognition can enhance user experiences and provide actionable insights. Despite progress in standalone deep learning models, limitations persist in capturing both spatial and temporal dependencies effectively. To address this, we developed a hybrid deep learning model combining Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks. CNNs were employed to extract spatial features from multivariate time-series sensor data, while LSTMs captured temporal dependencies crucial for accurate classification. The UCI HAR dataset, consisting of six activity labels, was used to benchmark the model. The implementation was carried out in Python, leveraging libraries like TensorFlow and Keras. The proposed hybrid CNN-LSTM model achieved an overall accuracy of 94%, with precision, recall, and F1-scores (macro and weighted averages) also reaching 94%. Individual activity labels recorded F1-scores ranging from 85% to 100%, demonstrating the model`s robustness across diverse activities. These findings validate the effectiveness of the hybrid CNN-LSTM architecture in overcoming the limitations of standalone models. The ability to capture both spatial and temporal patterns in sensor data underscores the model`s potential in advancing HAR applications. This study provides a foundation for future research in refining hybrid approaches, exploring additional datasets, and deploying such models in real-world applications. The results have significant implications for improving healthcare monitoring, fitness tracking, and human-computer interaction systems.