Data-Driven Prediction of Traffic Violations Using Improved Random Forest

Traffic violations represent a critical issue for public health and safety globally. This paper introduces a novel machine learning framework designed to predict traffic violations using comprehensive data collected over one year from Qatar. The dataset includes five key features: date, time, zone number, street number, and type of license plate, with the type of traffic violation as the target variable. After data cleaning and preprocessing, including the conversion of input data to numerical values and normalization, the class imbalance is addressed. Various sampling techniques are tested, with random oversampling yielding the best results. This study evaluates several machine learning models, including Naïve Bayes, Logistic Regression, Support Vector Machine (SVM), XGBoost, Neural Network, and Random Forest, for predicting traffic violations. To enhance performance, the proposed method introduces an improved version of the conventional Random Forest by incorporating novel modifications at two critical stages: tree construction and model evaluation. These enhancements aim to increase the accuracy and robustness of predictions, demonstrating the effectiveness of the approach over baseline models. Among these, the improved Random Forest model achieves the highest prediction accuracy at 99.5%. Feature importance analysis reveals that the street number is the most significant predictor, while the type of license plate is the least significant. The experimental results validate the effectiveness of the proposed method, highlighting its potential as a valuable tool for predicting and preventing traffic violations. This study aims to enhance traffic safety measures and demonstrates promise for broader applications in traffic management and safety systems.