SHORT TERM TRAFFIC FLOW PREDICTION USING MACHINE LEARNING - KNN, SVM AND ANN WITH WEATHER INFORMATION
Ibna Faysal, Rahman, A Al-Anazi, I Gates, S Bajaw, E Chung, M Kuwahara, L Cao, F Tay, S Clark, S; Datla, S Sharma, G Davis, N Nihan, M Dougherty, A Hasnat, F Rahman, J Hua, A Faghri, S Huang, B Ran, T Maze, M Agarwai, G Burchett, K Mller, A Smola, G Rtsch, B Schlkopf, J Kohlmorgen, V Vapnik, A Nagy, V Simon, F Rahman, A Hasnat, A Lisa
Abstract
Short term traffic prediction is one of the attractive topics in current Intelligent Transport System (ITS) research and practice. The rapid progress in machine learning and the emergence of new data sources makes it possible to observe and predict traffic conditions in cities more accurately than ever. In this study, three different sets of weather information or condition related parameters are incorporated with the different algorithms of machine learning for better traffic flow prediction. Three methods of machine learning-k-nearest neighbor (KNN), support vector machine (SVM), artificial neural network (ANN) are used in this research. However, it is hard to select the most appropriate machine learning traffic flow prediction model considering particular observation data. This paper shows the effect of the selection of each fundamental component of three machine learning algorithms and their effect on prediction accuracy. Five months of historical traffic flow data are trained with the weather condition. Then considering weather conditions, the traffic flow of one month is predicted. One hour interval traffic flow prediction considering weather information KNN gives more accurate results than SVM and ANN with 14.384% mean absolute percentage error and 0.948 R-square value.