Addressing Imbalanced EEG Data for Improved Microsleep Detection: An ADASYN, FFT and LDA-Based Approach

Hasan, Md Mahmudul; Khandaker, Sayma; Sulaiman, Norizam; Mahfuj Hossain, Mirza; Islam, Ashraful · 2024 · Crossref

DOI: 10.24237/djes.2024.17304

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 microsleep detection in drivers, focusing on the challenge of imbalanced electroencephalogram (EEG) data. Microsleeps are brief, involuntary lapses in consciousness that significantly increase accident risk, particularly during monotonous tasks like driving. While machine learning offers potential for automated detection, existing research often fails to adequately handle class imbalances in EEG datasets, leading to biased models that favor majority classes and misclassify minority microsleep events. The authors propose a novel framework combining Adaptive Synthetic Sampling (ADASYN), Fast Fourier Transform (FFT), and Linear Discriminant Analysis (LDA) to improve detection accuracy and robustness. The methodology utilizes a publicly available EEG dataset comprising recordings from 76 healthy individuals (50 men, 26 women). Data was collected using RemLogic devices at a 200 Hz sampling rate, with two EEG channels (O1-M2 and O2-M1) filtered via high-pass (0.3 Hz), low-pass (70 Hz), and notch (50 Hz) filters. The dataset was divided into wake, microsleep episode (MSE), microsleep episode with correction (MSEc), and eyes closed (ED) states. To address class imbalance, the ADASYN algorithm generated synthetic samples for minority classes, prioritizing difficult-to-learn instances. Feature extraction was performed using FFT to transform time-domain signals into frequency-domain matrices. Finally, LDA was employed for classification, reducing dimensionality by maximizing between-class scatter and minimizing within-class scatter. The dataset was split into 70% for training and 30% for testing. The experimental results demonstrate the effectiveness of the proposed approach. By applying ADASYN to balance the dataset, the model significantly reduced classification errors associated with imbalanced data. The framework achieved a peak testing accuracy of 92.71%. Additional performance metrics included a precision of 92.32%, recall of 92.05%, F1-score of 92.18%, and a Cohen’s Kappa score of 90.26%, indicating almost perfect agreement. The study confirms that addressing data imbalance through ADASYN, combined with FFT-based feature extraction and LDA classification, yields superior performance compared to methods using unbalanced data. The significance of this research lies in its contribution to reliable, non-invasive driver alertness monitoring systems. By effectively handling imbalanced EEG data, the proposed framework enhances the detection of microsleep episodes, which is crucial for preventing accidents caused by drowsy driving. The high accuracy and robustness of the model suggest its potential for practical implementation in real-world safety applications, offering a promising solution for mitigating risks in high-stakes professions requiring sustained attention.

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
• microsleep