LAUSR

Received: 6 September 2021 Accepted: 2 October 2021 Nowadays, many studies have been conducted to assess driver fatigue, as it has become one of the leading causes of traffic crashes. However, with the use of advanced features and machine learning approaches, EEG signals may be processed in an effective way, allowing fatigue to be detected promptly and efficiently. An optimal channel selection approach and a competent classification algorithm might be viewed as a critical aspect of efficient fatigue detection by the driver. In the present framework, a new channel selection algorithm based on correlation coefficients and an ensemble classifier based on random subspace k-nearest neighbour (k-NN) has been presented to enhance the classification performance of EEG data for driver fatigue detection. Moreover, power spectral density (PSD) was used to extract the feature, confirming the presented method's robustness. Additionally, to make the fatigue detection system faster, we conducted the experiment in three different time windows, including 0.5s, 0.75s, and 1s. It was found that the proposed method attained classification accuracy of 99.99% in a 0.5 second time window to identify driver fatigue by means of EEG. The outstanding performance of the presented framework can be used effectively in EEGbased driver fatigue detection.