Automated Sleepiness Detection via EEG Brainwave Analysis: A Nonlinear Ensemble Approach with Optimized Hyper-Tuning Strategies

K. Srujan Raju · 2024 · Crossref

DOI: 10.52783/anvi.v28.2005

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 critical safety issue of driver sleepiness, which contributes significantly to traffic accidents and fatalities globally. The research aims to enhance automated sleepiness detection systems by evaluating the performance of various machine learning algorithms on Electroencephalography (EEG) brainwave data. Specifically, the authors investigate whether hyperparameter tuning can significantly improve the accuracy and reliability of ensemble learning models in classifying drivers as awake or sleepy. The methodology utilizes a dataset obtained from Kaggle, comprising 3,735 EEG signal samples collected from four drivers using a single-channel NeuroSky MindWave headset. The data, recorded from the frontal lobe, was split into a training set of 2,614 samples and a test set of 1,121 samples. Five machine learning classifiers were employed: LightGBM, XGBoost, CatBoost, Extra Trees Classifier (ETC), and Random Forest (RF). The study first evaluated baseline performance without tuning, then applied grid search hyperparameter optimization to each model. Key hyperparameters tuned included learning rates, tree depths, number of estimators, and regularization terms, with 5-fold cross-validation used to determine optimal configurations. The results demonstrate that hyperparameter tuning substantially improved performance across all models. Before tuning, CatBoost achieved the highest baseline accuracy of 80%, followed by LightGBM and XGBoost at 79%, ETC at 78%, and RF at 77%. After optimization, CatBoost emerged as the superior model, achieving an accuracy of 83%, a recall of 78%, precision of 81%, and an F1-score of 0.80. Its Area Under the Curve (AUC) increased to 0.89, indicating strong discriminatory power between alert and drowsy states. While CatBoost required a longer prediction time of 2.78 seconds compared to the faster LightGBM (0.53 seconds) and XGBoost (0.58 seconds), its enhanced accuracy and robustness made it the most effective classifier. The tuned CatBoost model also showed improved Cohen’s Kappa and Matthews Correlation Coefficient values, confirming better agreement with actual labels. The study concludes that advanced machine learning techniques, particularly when combined with rigorous hyperparameter tuning, offer significant potential for improving automated sleepiness detection systems. CatBoost’s superior performance highlights its suitability for this domain, despite higher computational costs. The findings suggest that such optimized models can contribute to safer driving environments and improved worker safety by providing reliable, real-time detection of drowsiness. The authors recommend further exploration of hybrid approaches to achieve even greater stability and performance in practical applications.

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