Enhanced K-Nearest Neighbor Model For Multi-steps Traffic Flow Forecast in Urban Roads

Abstract

Short-term flow forecast is a fundamental key in intelligent transportation planning. Often accurate predictions are provided by the predictive models the most adapted to the nature of the addressed problem. In this paper we present a k-Nearest Neighbor approach (E-KNN) enhanced by taking advantage of traffic attributes. The proposed model is applied to 11 weeks of non-processed data, recorded by 7 inductive loop detectors installed on urban roads located in downtown of Bremen (Germany). The performance of E-KNN is tested on 3 weeks of data and reported following different day-hours categories, including rush hours. Excluding early day-hours where traffic is insignificant, E-KNN performs 6-steps (1h) prediction with an average absolute relative error of 17% on test-set.