Enhanced Multi-Class Driver Behavior Detection in IoMT Environments Using Hybrid LSTM-GRU Model

Abstract

The global rise in traffic accidents is significantly attributed to driver distractions —as a primary safety motivator—that can lead to collisions, injuries, fatalities, and property damage. Distractions such as using cell phones, eating, drinking, and interacting with in-vehicle systems are critical factors contributing to unsafe driving conditions. Addressing this issue necessitates advanced, real-time monitoring systems that can accurately detect and analyze driver behaviors. In this paper, we introduce an enhanced framework for real-time driver behavior detection within the Internet of Medical Things (IoMT) environment— i.e., a wearable-sensor edge platform compliant with IoMT architectures—that utilizes bio-signal data from wearable sensors. Our proposed system integrates Shimmer3 EMG sensors and a Raspberry Pi device to collect and process data, employing a hybrid Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models for classification. This framework includes three key modules: the data acquisition module, the hybrid LSTM-GRU training module, and the real-time analysis and feedback module. Experimental evaluations conducted on two real-life public datasets demonstrate that our hybrid model achieves a classification accuracy of 83.40% and 77.90% for the first and second datasets, significantly improving the detection of non-aggressive and aggressive driving behaviors. These results underscore the effectiveness of combining LSTM and GRU layers in capturing complex temporal dependencies, offering a scalable and reliable solution for enhancing road safety through improved driver behavior detection.