Traffic Congestion Prediction using Supervised Machine Learning Algorithms

Abstract

Traffic congestion is one of the main severe issues in many cities of the world. Monitoring and understanding traffic congestion is difficult because of its complex nature. The paper analyzed the efficiency of three supervised machine learning algorithms on traffic dataset that was downloaded from Kaggle repository. The dataset consists of 15 attributes and 33750 instances which were further divided into 70% for training and 30% for testing. The dataset was used to formulate predictive models for traffic congestion using three supervised machine learning algorithms: Classification Tree, Support Vector Machine and Ensemble (RUSBoosted) algorithms. The formulation and simulation of the predictive model were carried out using Matrix Laboratory (MATLAB) statistical tool. The results show Classification Accuracy (%) of 99.8, 99.9, 53.9. Prediction Speed (Obs/Sec) of 410000, 10000, 23000 and AUC of 0.66, 0.69, 0.91 for Classification Tree, Support Vector Machine and Ensemble (RUSBoosted) algorithms respectively.
The three models were compared and the best model in terms of accuracy was selected and validated. The study revealed that Support Vector Machine model has higher accuracy, followed by classification tree and Ensemble (RUSBoosted) algorithms. The model is recommended for transport Network services and any other machine learning algorithms can be used for traffic congestion predictive model.