EEG-Based Drowsiness Analysis to Establish Driving Safety Using Deep Learning Approaches

s, priyanka; s, shanthi · 2022 · Crossref

DOI: 10.21203/rs.3.rs-1491806/v1

archive: archived pipeline: cataloged

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

Summary

This paper addresses the critical safety issue of driver drowsiness, a leading cause of road accidents, by proposing a novel deep learning framework for real-time drowsiness detection using Electroencephalogram (EEG) signals. The motivation stems from the limitations of existing methods, which often suffer from high computational complexity, intrusiveness, or poor generalization across different subjects. The study aims to bridge these gaps by developing a model that efficiently learns long-term feature dependencies in EEG time-series data to predict the transition from an alert to a drowsy state. The methodology employs a stacked architecture combining a Deep Fully Connected-Convolutional Neural Network (DFC-CNN) with a stacked Long-Short Term Memory (s-LSTM) network. The DFC-CNN utilizes dilated convolutions to expand the receptive field for capturing temporal context, while the s-LSTM processes these features to extract long-term temporal dependencies. The experimental design utilized the OpenBCI sleep analysis dataset, comprising EEG data from 52 subjects (with 38 valid subjects used for final evaluation). Preprocessing involved butterfly filtering (1–60 Hz), Fast Independent Component Analysis (Fast ICA) for artifact removal, and wavelet thresholding for signal reconstruction. The model was trained and validated in a Google Colab environment using an 80/20 train-test split and k-fold cross-validation (k=3). The results demonstrate that the proposed s-LSTM with DFC-CNN model achieves an average prediction accuracy of 94.01%, with an average specificity of 95.69% and sensitivity of 94.40% across 20 test subjects. The model outperforms existing approaches, including SVM (59.60% accuracy), VMD-RIM (67.95%), and standard CNN (92.84%). Specifically, the configuration with 128 and 256 LSTM nodes yielded the highest performance. The model also showed superior generalization capabilities when validated on combined feature subjects. Additionally, the study analyzed Root Mean Square Error (RMSE) and scaling coefficients, finding that an alpha value of 0.8 resulted in the lowest RMSE (1.16) and a scaling correlation of 0.62. The significance of this work lies in its ability to provide a robust, non-intrusive, and computationally efficient tool for monitoring driver fatigue. By leveraging the strengths of both CNNs for feature extraction and LSTMs for sequence modeling, the proposed approach offers a better trade-off in accuracy and generalization compared to traditional machine learning and single-network deep learning models. This advancement supports the development of safer Brain-Computer Interface systems for automotive applications, potentially reducing accident risks by enabling timely interventions when drowsiness is detected.

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.

• drowsiness detection algorithms
• drowsiness

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