A Hybrid CNN-LSTM Deep Neural Network Model for Efficient Human Activity Recognition

This is a comprehensive exploration of Human Activity Recognition (HAR) and its integration with deep learning techniques, specifically Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks. CNNs are widely used in HAR by extracting spatial features; however, CNNs are not competent of extracting global features. LSTM network evaluates sequence data by looping across time steps, and identify long-term dependencies between time steps, which plays a crucial part in HAR. To address this concern this study primary objective is to develop a CNN-LSTM deep neural network-based HAR model. The paper begins with an introduction outlining the objectives and constraints of the research. A detailed literature review follows, delving into the fundamentals of HAR, its relationship with machine learning, and the introduction of CNN and LSTM in this context. The methodologies section covers data gathering, preprocessing, model architecture, hyperparameter tuning, and the learning and evaluation processes. The challenges within the field of HAR, such as intraclass differences, interclass similarity, and recognition in real-world scenarios, are discussed in the subsequent section. The results of the model execution, including the performance of the CNN + LSTM model and a comparison with baseline models, are presented and discussed. This paper concludes by highlighting the key findings and their real-life use case scenarios, and it discusses the potential for future research in the field of HAR. In summary, this paper aims to contribute to the understanding and advancement of HAR models, particularly in complex, diverse, and real-world scenarios. The proposed methodology enhances the recognition accuracy with value of 98.67% and F1 score as 95.84%.