VANET Severity Classification in BSM Messages with a Novel Naïve Bayes Feature Selection

Concerning traffic safety, the vehicle ad-hoc network (VANET) has a protocol for exchanging messages between vehicles called Basic Safety Message (BSM), which prioritizes communication-based on severity. Several studies have existed in classifying the severity of an event message, but classifying severity using machine learning and a novel naïve Bayes feature selection is a research opportunity. This research synthesizes a classification methodology with machine learning with severity classification in BSM for VANET. We use datasets from previous research, which were obtained based on simulations. Then we carry out pre-processing, including a feature selection process with the novel multinomial naïve Bayes features selection (MNBFS), standardization, and oversampling. We designed four machine learning models to compare: decision tree, SVM, naïve Bayes, and KNN. Finally, we reached the four methods with the receiver operating curve (ROC), the area under the curve (AUC), and the compression ratio (CR). The test results show that SVM is the machine learning model with the best performance. Then the normalization process can improve SVM performance. The SVM AUC value for severity classification in VANET is 0.986. We also reveal the findings of our novel feature selection method. The MNBFS can be a model compression for severity classification in VANET on fog computing, with CR = 1.1×.