Toward Automatic Vehicle Classification for Toll Booths on Chilean Interurban Highways

This work addresses the challenge of improving the accuracy of vehicle classification at toll booths on Chilean interurban highways through the application of machine learning techniques. The study focuses on a highway in the Bío Bío region, where a set of representative data is collected and prepared, including detailed information on vehicles registered and classified by the concessionaire company. The total data used consists of 961,915 records. Four machine learning models are evaluated: K-Nearest Neighbors (KNN), Multilayer Neural Network (MLP), Support Vector Machine (SVM), and Decision Tree Classifier (DTS). The models are trained and evaluated using standard metrics such as accuracy, precision, recall, and F1-Score. Additionally, confusion matrix, learning curve, ROC curve, and Precision-Recall curve are assessed. The experiments show that KNN and MLP are the models with the best results, achieving an accuracy greater than 98.6% and an F1-Score close to 97.9%, thereby improving the accuracy of the current toll booth system and exceeding the 95% accuracy required by the Chilean Ministry of Public Works.