Using neural network for drowsiness detection based on EEG signals and optimization in the selection of its features using genetic algorithm

Sarabi, Sepehr; Asadnejad, Milad; Rajabi, Saman · 2020 · Crossref

DOI: 10.15649/2346075x.1004

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 drowsiness, a leading cause of traffic accidents responsible for over 20% of road casualties globally. Motivated by the need to reduce traffic deaths, the research proposes a non-invasive detection system using electroencephalography (EEG) signals to determine whether a driver’s eyes are open or closed. This approach avoids the privacy and distraction concerns associated with camera-based monitoring systems. The primary objective is to develop an efficient neural network model capable of classifying EEG data with high accuracy while optimizing computational resources for potential microcontroller implementation. The methodology involves collecting EEG data from a statistical population of 600 individuals using an Emotive EEG neuroheadset. The dataset comprises 14 features derived from brain channel leads, recorded over 117 seconds, and categorized into two classes: open eyes and closed eyes. The study employs two types of artificial neural networks for classification: the Perceptron neural network and the Radial Basis Function (RBF) network. To enhance performance, the researchers optimized the selection of features and the fitting function using a genetic algorithm. This optimization aimed to improve the Fisher’s discriminant rate, thereby reducing execution time and memory space requirements. The Perceptron network was specifically tested with different transform functions, including Tansig and Logsig, and varying numbers of neurons to evaluate their impact on classification accuracy. The results demonstrate that the neural network models effectively classify the EEG signals. For the Perceptron network using the Tansig transform function, the accurate classification rates varied slightly with the number of neurons: 98.70% for 10 neurons, 99.25% for 20 neurons, 98.48% for 30 neurons, and 99.04% for 40 neurons. The application of the genetic algorithm for feature optimization yielded improved results in terms of both speed and space efficiency compared to unoptimized models. The study confirms that EEG-based detection, combined with optimized neural networks, provides a reliable method for distinguishing between alert and drowsy states. The significance of this work lies in its contribution to the development of compact, efficient drowsiness detection systems for automotive safety applications. By utilizing genetic algorithms to optimize feature selection and neural network parameters, the study achieves high classification accuracy while minimizing computational load. This makes the proposed system suitable for integration into resource-constrained microcontrollers, facilitating real-time monitoring of driver alertness. The findings support the viability of EEG signal processing as a robust alternative to visual monitoring, offering a non-intrusive solution to mitigate drowsy driving accidents.

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