Forecasting the Subway Volume Using Local Linear Kernel Regression

Abstract

Entrusted by the Kaohsiung Rapid Transit Corporation (KRTC), this study attempts to devise a more effective methodology to forecast the passenger volume of the subway system in the city of Kaohsiung, Taiwan. We propose a local linear kernel model to incorporate different weights for each realized observations. It enables us to capture richer information and improve rate of accuracy. We compare different methodologies, for example, ARIMA, Best in-sample fit ARIMA, linear model, and their rolling versions with our proposed local linear kernel regression model by examining the in-sample and out-of-sample performances. Our results indicate that the proposed rolling local linear kernel regression model performs the best in forecasting the passenger volume in terms of smaller prediction errors in a wide range of measurements.