Simulation of Autoregressive Integrated Moving Average- Generalized Autoregressive Conditional Heteroscedasticity (ARIMA-GARCH) to Forecast Traffic Flow

Abstract

Modeling the unprecedented traffic flow data generated by Intelligent Transportation Systems can boost the innovation-capacity of the transportation management systems to drive informed decision-making. Thus, this paper attempts to simulate traffic forecasting techniques that can be adopted in the Philippines to make fact-based decisions into accurate and effective traffic management schemes. In this research, a schematic framework is introduced organized into three stages (Preprocessing, Model Identification and Estimation, and Model Checking) sequentially arranged to comprehensively estimate the best-appropriate model to forecast traffic flow using ARIMA and GARCH models. The Model Identification and Estimation is the conditional stage in the framework that pre-determines if hybrid modeling is necessary based on the given datasets. Various accuracy metrics are also used to find the “best” model and select the optimal values for ARIMA and GARCH models. The proposed framework is simulated in R Programming using the vehicular traffic flow datasets at North Avenue, EDSA northbound, Manila, Philippines. The resulting models, consist of the best fit ARIMA (1,1,3) and GARCH (1,2), are combined as the hybrid model and compared using its prediction results. Based on the visual simulation data, the prediction accuracy result of the ARIMA model outperforms the combined ARIMA-GARCH model given the actual data. Conclusively, the simulation performance provides proof to suggest that the forecasting models are timely tools to predict future traffic flow and aid in making better traffic inventions and schemes.