Enhancing Road Safety with Smartphone-Based Machine Learning Driver Behavior Classification and Aggressive Driving Detection Using Feature Reduction Methods

Drivers’ behaviors are receiving more attention because of the disturbingly frequent rate of vehicle collisions. In road safety reports, human behavior is typically recognized as the most significant indicator of accident probability. The identification and classification of aggressive or deviant driving behavior is a critical necessity in the real world for preventing fatal traffic accidents and protecting other road users. In addition to assisting in the implementation of corrective steps, automatic identification of a driver's behavior aids in the prevention of potentially dangerous scenarios for the driver and all other participants in the driving environment. In this study, we suggested a multi-stage system for detecting aggressive behavior. The first stage involves splitting the driving behavior dataset into a training and testing set. The second stage is the pre-processing stage, which is critical for preparing the data for the following stage, and we employed a normalization technique. Then, in the following stage, two approaches will be taken. The first strategy is to directly enter the data into the three classifiers utilized in this work to classify the driving behavior, and the second approach is to choose features using three of the most well-known approaches for doing so. The next stage is classification, which is performed using six ML algorithms (KNN, NB, SVM, LR, SGD, and AdaBoost). The results obtained indicate that the AdaBoost algorithm outperforms the rest of the classifiers in all cases, as the detection accuracy reached 82% without feature selection, and with PCA, SVD, and MI, but with SVD-5 the detection system was in the fastest cases.