Abstract
This study proposes an integrated machine learning framework for advancedtraffic analysis, combining time-series forecasting, classification, andcomputer vision techniques. The system utilizes an ARIMA(2,0,1) model fortraffic prediction (MAE: 2.1), an XGBoost classifier for accident severityclassification (100% accuracy on balanced data), and a Convolutional NeuralNetwork (CNN) for traffic image classification (92% accuracy). Tested ondiverse datasets, the framework outperforms baseline models and identifies keyfactors influencing accident severity, including weather and roadinfrastructure. Its modular design supports deployment in smart city systemsfor real-time monitoring, accident prevention, and resource optimization,contributing to the evolution of intelligent transportation systems.