EEG-based Negative Emotion Classification while Self Driving in a Simulated Environment

Abd Halin, Abdul Hafiz; Wan Ahmad, Wan Khairunizam; Omata, Masaki; Kamarudin, Latifah Munirah · 2026 · Crossref

DOI: 10.58915/ijaris.v2i1.2852

archive: archived pipeline: cataloged verified

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

Summary

This study addresses the challenge of accurately classifying driver emotions, specifically negative states, during autonomous driving scenarios. The research is motivated by the need for improved affective computing and driver monitoring systems that can objectively detect emotional responses to hazardous situations. While subjective methods like the Self-Assessment Manikin (SAM) provide psychological grounding, they are prone to bias and variability. The authors propose a framework combining SAM for validation with electroencephalography (EEG) signal analysis and machine learning to achieve reliable, real-time emotion classification. The methodology employed a two-stage experimental design. First, 34 participants completed SAM questionnaires after observing autonomous driving scenario videos to validate that specific scenarios elicited target emotions. Second, 18 participants engaged in an immersive simulated driving environment involving a steering wheel and pedals, with EEG signals recorded via an Emotiv Epoc X device. Participants experienced four situations: a baseline training phase, sudden braking by a lead car, a dangerous lane change by a nearby vehicle, and a pedestrian near-miss. EEG signals underwent preprocessing using a 6th-order bandpass filter and zero-phase distortion filtering to remove artifacts, followed by normalization. Feature extraction involved five time-domain features (mean, standard deviation, skewness, kurtosis, entropy) and three frequency-domain features (absolute, relative, and peak power across delta, theta, alpha, beta, and gamma bands). Mutual Information (MI) was used for feature selection, reducing the set to the two most discriminative features: entropy and standard deviation. The study evaluated two classification algorithms: k-Nearest Neighbors (KNN) and Random Forest (RF), using a 70/30 train-test split and 5-fold cross-validation. The KNN classifier achieved an overall accuracy of 98%, with perfect precision and recall for relax, focus, and nervous states, and an F1-score of 0.95 for fear and surprise. The RF classifier achieved 93% accuracy, performing strongly on fear (F1-score 1.00) and nervousness (F1-score 0.90) but showing lower performance on focus and surprise compared to KNN. The results indicate that KNN was more effective in this context, likely due to the compact, well-separated feature space created by the MI-based selection, whereas RF’s ensemble advantages were constrained by the limited feature diversity. The findings demonstrate that EEG-based classification, particularly using entropy and standard deviation, can reliably distinguish between emotional states such as fear, surprise, nervousness, focus, and relaxation. The high accuracy of KNN suggests its suitability for real-time driver monitoring applications due to its low computational complexity. The study concludes that while SAM provides necessary experimental validation, EEG offers an objective, continuous measure of driver emotion. The authors acknowledge limitations, including the small sample size, homogeneous demographic of university students, and the simulated nature of the driving environment, suggesting future work should expand to real-world conditions and larger, more diverse datasets.

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.

• 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, validation psychometrics