An application of intelligent noise filtering techniques in demand forecasting for carsharing systems

Abstract

This study deals with demand forecasting for the carsharing system with multi-station and flexible returning station and time. Neural network is employed as the simulation model to forecast the demand at each station at certain time period in the system. This study distinguished from the previous study because it introduces intelligent filtering techniques as a tool to remove the noise of the data before it is fed into the simulation model. Two filtering techniques have been tested, namely outlier analysis and cluster analysis. Results show that this extra procedure helps to enhance the forecasting model as well as improve the forecasting accuracy.