An Innovative Approach on Driver's Drowsiness Detection through Facial Expressions using Decision Tree Algorithms

This study aims to incorporate detecting driver drowsiness through the use of facial expressions as the basis. Instead of using a driving simulator, real-life driving of 10 drivers was our approach in conducting the experiment. Also, all drivers were driving through their normal routine using a sedan-type car, mostly having passengers beside them since the goal is to obtain drowsiness in a real-life driving situation. Furthermore, this study focuses only on classifying three drowsiness levels: "No Drowsiness", "Mild Drowsiness "and "Extreme Drowsiness". This study showed that combining Action Units for eye closure and facial expressions such as eyebrows was essential in this aspect. Results show that Random Forest classifier using CART algorithm was the most suitable model. The model performance was further improved by combining eye closure and facial expression features defined by the Chi Square attribute evaluator and by removing some features that were ranked insignificant.