Title: Carsharing customer demand forecasting using causal, time series and neural network methods: a case study
Authors: Elnaz Moein; Anjali Awasthi
Addresses: Faculty of Engineering and Computer Sciences, Concordia University, CIISE, Montréal, Canada ' Faculty of Engineering and Computer Sciences, Concordia University, CIISE, Montréal, Canada
Abstract: Carsharing services are becoming popular in recent times. Deploying right number of fleet at stations is a critical component in assuring high quality service for customers. This can be done efficiently if customer demand is predictable or known in advance. In this paper, we address the problem of customer demand forecasting for improving carsharing operations. Three categories of methods namely causal (regression forecast, regression forecast with seasonality adjustments), time series (exponential smoothing, moving average) and neural networks are evaluated for forecasting customer demand. An application of the proposed methods on demand data from a carsharing organisation called Communauto is provided. The results of our study show that neural network is the best method in this prediction. The proposed work has strong practical applicability. Having an accurate forecast of the customers' demands in different times of the year can help increase customer satisfaction and reach business performance targets. Especially if electric vehicles are used in carsharing companies, since they require special infrastructures.
Keywords: demand forecasting; carsharing; regression; exponential smoothing; moving average; neural networks.
International Journal of Services and Operations Management, 2020 Vol.35 No.1, pp.36 - 57
Received: 17 Nov 2016
Accepted: 29 Oct 2017
Published online: 02 Jan 2020 *