Road rage is a critical factor in traffic accidents, often expressed through vocal and semantic cues. This study proposes a multimodal road rage detection system that integrates speech and text features. Speech signals are processed by a probabilistic neural network (PNN) optimized with an Improved Sand Cat Swarm Optimization (ISCSO) algorithm, while text features are modelled with a long short-term memory (LSTM) network. A decision-level fusion strategy combines outputs from both modalities. Experiments on a self-constructed dataset of 10,000 speech samples and corresponding text corpora demonstrate that the proposed model achieves superior performance compared to CNN, DBN, TextCNN, and hybrid deep learning baselines. The system achieved maximum accuracy and recall of 95.61% and 99.31%, while maintaining low computational overhead (minimum detection time 58 ms, memory usage 10.06%). These findings suggest that the ISCSO-PNN and LSTM multimodal fusion framework provides an efficient and effective approach to detecting road rage, with strong potential for integration into real-time intelligent transportation systems.
Xie et al. (Sun,) studied this question.