k-Nearest Neighbor Model for Multiple-Time-Step Prediction of Short-Term Traffic Condition

Abstract

One of the most critical functions of an intelligent transportation system (ITS) is to provide accurate and real-time prediction of traffic condition. This paper develops a short-term traffic condition prediction model based on the k-nearest neighbor algorithm. In the prediction model, the time-varying and continuous characteristic of traffic flow is considered, and the multi-time-step prediction model is proposed based on the single-time-step model. To test the accuracy of the proposed multi-time-step prediction model, GPS data of taxis in Foshan city, China, are used. The results show that the multi-time-step prediction model with spatial-temporal parameters provides a good performance compared with the support vector machine (SVM) model, artificial neural network (ANN) model, real-time-data model, and history-data model. The results also appear to indicate that the proposed k-nearest neighbor model is an effective approach in predicting the short-term traffic condition.