Abstract
ABSTRACT Human Activity Recognition (HAR) is essential for allowing assistive services in automated residential environments. Existing methods still have limitations, including handling complex action sequences, processing costs from high-dimensional sensor data, and collecting long-term connections. To address these issues, this paper presents a hybrid Deep Residual Bi-LSTM Network (ResBi-LSTM). This approach efficiently extracts deep features and captures temporal context by combining the advantages of residual networks, Bidirectional LSTM architectures, and one-dimensional convolution layers. Enhanced Deep Residual blocks for feature extraction, repeated residual network blocks with skip connections for improved learning, and Bi-LSTM layers for capturing long-term patterns are all included in the ResBi-LSTM model. In the present research, the efficacy of the suggested network has been assessed with an accuracy of 95.07%, precision of 95.27%, recall of 94.99%, and F1-score of 99.83% on HAR tasks in smart homes. Based on experimental results, the ResBi-LSTM model is an appropriate choice for assistive technology and smart home systems in real-life situations since it can effectively identify human behaviors while reducing computational complexity.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call