Random Forest Algorithm-based Multi-Feature Vector Optimization for Fatigue Driving Vigilance Monitoring

Accidents caused by fatigue account for a large proportion of traffic, so it has great significance in preventing fatigue driving by monitoring the vigilance levels of drivers. In order to monitor the changes of vigilance, an experiment was designed to collect the electroencephalogram (EEG) signals of drivers at different vigilance levels, and further to find out the major factors. Several commonly used features, EEG sub-bands and EEG channels were used, and the combination of which can not only improve the diversity of vigilance feature vectors, but also improve the classification accuracy of vigilance monitoring. However, too many features in the feature vector will result in a large amount of calculations. Therefore, this paper used the decision tree (DT) and random forest (RF) algorithm to optimize feature vector without affecting the classification accuracy of vigilance monitoring. DT was used to establish the model with all feature vectors, then by comparing the weight of features, EEG sub-bands and EEG channels, the optimized feature vector FP1-FP2/d4-d5-d6/CV was obtained. RF was used to calculate the classification accuracy with the optimized multi-feature vector to ensure the high monitoring accuracy of the changes of vigilance.