Enhancing the Performance of a Model to Predict Driving Distraction with the Random Forest Classifier

Abstract

Distracted driving is known to be one of the main causes of crashes in the United States, accounting for about 40% of all crashes. Drivers’ situational awareness, decision-making, and driving performance are impaired as a result of temporarily diverting their attention from the primary task of driving to other unrelated tasks. Detecting driver distraction would help in adapting the most effective countermeasures. To tackle this problem, we employed a random forest (RF) classifier, one of the best classifiers that has attained promising results for a wide range of problems. Here, we trained RF using the data collected from a driving simulator, in which 92 participants drove under six different distraction scenarios of handheld calling, hands-free calling, texting, voice command, clothing, and eating/drinking on four different road classes (rural collector, freeway, urban arterial, and local road in a school zone). Various driving performance measures such as speed, acceleration, throttle, lane changing, brake, collision, and offset from the lane center were investigated. Using the RF method, we achieved 76.5% prediction accuracy on the independent test set, which is over 8.2% better than results reported in previous studies. We also obtained a 76.6% true positive rate, which is 14% better than those reported in previous studies. Such results demonstrate the preference of RF over other machine learning methods to identify driving distractions.