A Machine Learning-Based Data Fusion Model for Online Traffic Violations Analysis

Abstract

AbstractTraffic violations occur due to driving or behavioral issues that result in traffic offense and violate the law. Traffic violations such as running red lights, speeding, and reckless driving are translated to millions of traffic infractions every year. This paper proposes a machine learning-based data fusion (MLDF) model for online traffic violations analysis (OTVA) system. The MLDF model is set to perform cumulative traffic analysis by using a software agent (SA) for decision making and Gradient Boosted Trees (GBT), Naive Bayes (NB), and Random Forest (RF) algorithms for classification. The MLDF model is incorporated in the OTVA system for categorizing traffic violation types online. The performance of the MLDF model that includes the SA and the NB, GBT, and RF algorithms is measured and compared in terms of accuracy, recall, precision, and f-measure. The results show that the MLDF model outperforms the single NB and RF algorithms in which GBT achieves 69.86% (±1 .28%) accuracy, NB achieves 66.02% (± 3.38%) accuracy, RF achieves 69.36% (± 0.84%) accuracy, and MLDF achieves 71.88% (± 1.23%) accuracy scores. It is hoped that the results of this paper can serve as a baseline for investigations related to the use of advanced models to automate the detection of traffic violations.KeywordsTraffic violationData fusionMachine learningPrediction software agentDecision making