Exploratory Study on Approaches for Traffic count Prediction; Using Toll-Way Traffic Count

Abstract

The process of predicting or simulating traffic conditions, based on current and past traffic observations is an important component of any of the Intelligent Transportation System (ITS) applications. There are several methods to predict traffic count. However, each method in different literatures may use different datasets, different time intervals of input traffic flow/count. One of the aims of this study is to provide a review and performance analysis of parametric and non-parametric approaches on traffic prediction. Second, to explore the possibilities in the implementation of Advanced Traffic Management Systems (ATMS), one of the functional areas of Intelligent Transportation Systems (ITS), by predicting traffic count on toll plazas for optimizing of toll-plaza operations. An RFID (Radio-Frequency Identification) based Electronic Toll Collection (ETC) system gives timely varying traffic counts observed at toll plazas, which has been utilized to develop prediction models based on historic data. An empirical differentiation of four methods, namely Seasonal Autoregressive Integrated Moving Average (SARIMA) model based on time series analysis, Monte Carlo Simulation (MCS), Random Forest (RF) based on tree ensemble learning technique and KNN non-parametric regression-based machine learning technique, are proposed. Performance analysis at varying time intervals (5, 10, and 15 minutes) of input traffic count, for all the aforesaid models were compared with Simple Average Technique (SAT) using the historic data collected from two different toll-plazas in India. It was observed that K Nearest Neighbors (KNN) non-parametric regression performed better than other methods in most of the cases.