CLASSIFICATION OF COGNITIVE WORKLOAD FROM EEG SIGNALS USING MULTIDIMENSIONAL FEATURES WITH MACHINE LEARNING AND DEEP LEARNING

Koca, Yavuz Bahadir · 2025 · Crossref

DOI: 10.21923/jesd.1669626

archive: archived pipeline: cataloged

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 cognitive workload levels from electroencephalogram (EEG) signals, a critical need in fields like aviation, education, and human-computer interaction where mental fatigue impairs decision-making. The research aims to determine if integrating multidimensional features—spanning time, frequency, and complexity domains—with machine learning (ML) and deep learning (DL) models can yield high-precision classification of low, medium, and high cognitive load states. The methodology utilized the Simultaneous Task EEG Workload (STEW) dataset, comprising EEG recordings from 45 university students (out of 48 initial participants) under resting and high-load multitasking conditions. Data were acquired using a 14-channel Emotiv EPOC system at 128 Hz. Preprocessing involved Butterworth band-pass filtering (1–50 Hz) and notch filtering to remove 50 Hz power-line interference. Signals were segmented into windows of 256 and 512 samples. A total of 309 features were extracted per window, including time-domain metrics (mean, variance, RMS, zero-crossing rate), frequency-band powers (delta, theta, alpha, beta, gamma), and complexity measures (Shannon entropy, Hjorth parameters, Hurst exponent, and Higuchi fractal dimension). These features were fed into five ML algorithms (Logistic Regression, SVM, k-NN, Random Forest, XGBoost) and three DL architectures (DNN, 1D-CNN, LSTM). Class imbalance was managed via weighted sampling and stratified splits. Results indicated that ML models outperformed DL models in this specific feature-engineering context. The XGBoost algorithm achieved the highest overall accuracy of 99.4% and a Cohen’s Kappa of 0.990. Among deep learning approaches, the LSTM model performed best with 98.75% accuracy and a Kappa of 0.981, effectively capturing temporal dependencies in the EEG sequences. Channel-based analysis revealed that EEG signal complexity, measured by fractal dimension and entropy, increased significantly in the frontal and temporal regions during high cognitive load, while amplitude increases were observed in frontal and left temporal channels. The significance of this work lies in demonstrating that a comprehensive, hybrid feature set derived from EEG signals enables near-perfect classification of cognitive workload when paired with gradient-boosted ML algorithms. The findings suggest that XGBoost is a robust choice for EEG-based cognitive state assessment, while LSTM remains a strong alternative for capturing time-series dynamics. Furthermore, the identification of specific brain regions (frontal/temporal) as key indicators of load provides neurophysiological insight that can inform the design of adaptive human-machine interfaces and fatigue monitoring systems.

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 qwen3.8-27b-gittensor on 2026-08-23; verification: pending re-verification.

Topics

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

• neuro workload indices
• 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