Prediction of Emergency Braking Intention using Machine Learning Models

Abstract

Since 2000, road accidents are on the rise, being a leading cause of death worldwide. Approximately 94% of all traffic crashes are due to human mistakes. These mistakes include speeding, reckless driving, or driving under the influence. A significant proportion of automobile accidents could be avoided with emergency braking support. Driver’s status monitoring and human mistake detection are some of the most successful applications of electroencephalogram (EEG) signals. This paper proposes a prediction model for predicting the intention of the driver to use emergency braking using the driver’s electroencephalogram (EEG) signals coupled with electromyography (EMG) data from leg muscles. The dataset utilized in this investigation was obtained from eighteen subjects while driving a simulated car by using an electrode cap with 64 scalp sites. The electroencephalogram (EEG) data signals are segmented to a 150 ms window and applied to five different machine learning classifiers (k-Nearest Neighbor, Support Vector Machine, Random Forest, Logistic Regression, and Naïve Bayes) for prediction. The proposed model can successfully predict the driver’s emergency braking intention 150 ms before the moment of the brake with an accuracy of 99.6%; that is, at 100 km/h driving speed, our model can anticipate emergency braking intention 4.22 m earlier. Furthermore, the model increased the driver’s prediction of emergency brake intention by 15.2% compared to other models.