Predicting stress levels of automobile drivers using classical machine learning classifiers

Abstract

Traffic congestion has been found to be a substantial source of stress for many people. To put it another way, driving under the influence of high amounts of stress could result in accidents and other long-term health issues. As a result, there is a pressing need to measure and anticipate the key reasons (features or classes) that raise stress levels in drivers. Car drivers’ stress levels while driving are investigated using four different data mining algorithms: K-Nearest Neighbor (KNN), Decision Tree, PART, and Naive Bayes. A wearable biomedical device attached to the driver’s body was used in an experiment in Amman, Jordan, where 12 drivers traveled different routes. According to the obtained data (dataset), ‘Yes’ indicates the presence of stress whereas ‘No’ indicates the absence of stress. Naïve Bayes surpasses the PART, KNN, and Decision Tree in terms of AUC and Precision, but the KNN is superior in terms of Recall and F1 for the other classifiers. Furthermore, the results show that the level of stress differs depending on the gender, age, driving skills, distraction, and driving-concentration features.