Evolutionary-Optimized Multimodal Driver RiskAssessment Using DE-Enhanced Cross-AttentionFusion of EEG and Telematics

Raga Madhuri, Ch; Manikonda, Dedeepya; Chinta, Madhurya · 2026 · Crossref

DOI: 10.5815/ijigsp.2026.04.10

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the limitations of unimodal driver monitoring systems, which often fail to capture the complex interplay between a driver’s internal cognitive-emotional state and external driving behavior. Motivated by the high global mortality rate from traffic accidents and the inadequacy of existing systems that rely solely on vehicle data or visual cues, the authors propose HECANet (Hybrid Evolutionary Cross-Attention Network). This multimodal framework integrates Electroencephalography (EEG) signals, which reflect internal cognitive dynamics, with vehicle telematics data, which captures observable driving behaviors, to provide a holistic assessment of driver risk. The methodology employs a four-module architecture. First, EEG signals from the DEAP dataset are preprocessed and modeled using a Convolutional Neural Network–Long Short-Term Memory (CNN–LSTM) architecture, with hyperparameters optimized via Particle Swarm Optimization (PSO) to capture spatiotemporal emotional patterns. Second, telematics data from the Levin Vehicle Telematics dataset are analyzed using an XGBoost classifier, optimized through a Genetic Algorithm (GA), to identify driving behaviors. Third, a Differential Evolution (DE)-optimized cross-attention fusion mechanism aligns and integrates the EEG and telematics embeddings, allowing each modality to attend to relevant features in the other. Finally, K-Means clustering categorizes the fused representations into Safe, Caution, and Risky groups. The study utilizes independent datasets for EEG and telematics, performing representation-level fusion rather than temporal alignment. Experimental results demonstrate that HECANet significantly outperforms baseline models and previous approaches. The optimized multimodal model achieved a test accuracy of 94.7% and a macro F1-score of 0.94. In comparison, unoptimized baseline models for EEG, telematics, and simple fusion achieved accuracies of 82%, 85%, and 88%, respectively. The evolutionary optimization strategy (PSO, GA, and DE) proved superior to grid search, which yielded 91.2% accuracy. Class-wise analysis revealed perfect precision and recall for the "Mild Risk" category, with strong performance for "Safe" (91.5% F1) and "High Risk" (91.1% F1) classes. Statistical validation across multiple runs confirmed model stability, with low standard deviations in accuracy and F1-scores. The significance of this work lies in its demonstration that joint emotion–behavior modeling, enhanced by evolutionary optimization, substantially improves driver risk prediction. By effectively capturing non-linear interactions between cognitive states and driving actions, HECANet offers a robust, interpretable solution for intelligent transportation systems and fleet safety applications. The study highlights the potential of multimodal learning to detect risk before it manifests in observable behavioral anomalies, addressing a critical gap in current driver monitoring technologies.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• telematics crash prediction
• drowsiness detection algorithms

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: physiological data
• Methodological Resource: tool software
• Theoretical Contribution: computational model