SHORT-TERM ROAD TRAFFIC FLOW PREDICTION UTILIZING TIME SERIES ANALYSIS METHOD CASE STUDY: CITY OF KURE, HIROSHIMA, JAPAN

• Main Author: Ramadhan, Rizqi
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/57199
• Institution: Institut Teknologi Bandung

Road traffic possesses a dynamic and anticipatory nature. Previous medium-long interval macro traffic models that require complex data are becoming irrelevant for dealing with the stochastic nature of traffic. Thus, short-term prediction approach come as a solution to model the dynamic and continuity of traffic. This research is structured to formulate short-term traffic flow prediction model alternatives, in which the best-performing model will be used to predict traffic flow at a downstream point on the National Route 31 arterial road, Kure City, Hiroshima, Japan. The models will utilize statistical time series analysis approach, consisted of ARIMA method and Dynamic Regression method. The modeling process resulted in 2 alternative ARIMA models and 10 alternative Dynamic Regression models. Based on point evaluation and distributional forecast accuracy, the Dynamic Regression of variable vu,t with ARIMA(0,0,6)(0,1,0) error was chosen as the model with the best accuracy. The capability of the selected model will be tested further by generating traffic flow predictions in 4 different prediction horizons outside the development environment. The selected Dynamic Regression Model successfully produces remarkable point forecasts and prediction intervals of downstream traffic flow in which noted by indicators; maximum bias of -6.96, the maximum Root Mean Square Error of 23.86, maximum Mean of Absolute Percentage Error of 12.53%, and maximum Continuous Ranked Probability Score of 13.89.