Traffic Congestion Prediction: A Machine Learning Approach

In this study, in order to forecast traffic flow, we employed univariable and multivariable regression models, including Linear Regression (LR), Random Forest (RF), Multi-Layer Perceptron (MLP), Gradient Boosting (GB), Long short-term memory (LSTM), and Gated Recurrent Unit (GRU). With timestamp information being unavailable, we use temporal characteristics and seasonality time gaps between timeseries data to improve the results. GB outperformed the other models with regards to the coefficient of determination ( \(R^2\) ), Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), as well as Explained Variance Score (EVS). One extra univariable approach was utilized, combining data from other highly correlated intersections. This strategy led to better models with regard to several metrics. All in all, the multivariable and highly correlated approach produced better results compared to univariable regression.