MACHINE LEARNING BASED TRAFFIC FLOW PREDICTION USING URBAN TRANSPORTATION DATA

Abstract

The spread of the city, rising vehicle population, and the mounting demand for transportation have contributed to making traffic a major challenge for the modern cities. The effective prediction of the traffic flow is a key to enhancing the efficiency of transportation and alleviating congestion and to aid intelligent transportation systems (ITS). The paper explores how machine learning can be used to forecast traffic flow patterns in urban transport systems by leveraging data analysis. Three different models were compared, the Support Vector Regression (SVR), the Random Forest (RF) and the Long Short-Term Memory (LSTM) models, by means of a developed comparative framework, which is based on traffic-related variables such as traffic volume, vehicle speed, vehicle occupancy rate, weather conditions and temporal information. Research is based on quantitative experimental design in the data collection process, while data preprocessing, feature engineering, model training, and performance evaluation are done. The Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE) and coefficient of determination (R²) were used to evaluate the accuracy of the prediction. The results show that the machine learning algorithms are capable of modelling complex traffic patterns and providing reliable traffic flow forecasts. However, LSTM has the best prediction performance among them, because LSTM can model the long-term temporal dependency and nonlinear relationship among traffics. Random Forest had a good performance and was good for interpretability whereas SVR had scoring with acceptable accuracy in highly dynamic traffic condition. The study also pinpointed the maximum effect on the traffic prediction as caused by historical traffic volume, traffic occupancy rate, and temporal variables. The findings demonstrate the potential for machine learning-based forecast systems to be used to improve traffic management, support data-driven decision making and urban mobility. Moreover, this work has a significant role in the newly emerging “smart transportation” research area, in particular with respect to the development of scalable and reliable traffic forecasting models for sustainable urban transportation planning and intelligent transportation services.