SPATIO-TEMPORAL MODELS FOR FORECASTING FOOD DELIVERY DEMAND

Abstract

Many real industry problems involves spatio-temporal demand prediction, which requires predicting demand at a certain time across different locations. Most of the demand data are not continuous variables but are instead count variables, like the number of orders and number of transactions. Previous research studies have mainly implemented neural network methods for spatio-temporal problems. In this paper however, we would explore more interpretable methods and hence will focus on how we can apply classical time series models, such as Vector Autoregressive (VAR), Spatio-Temporal Autoregressive (STAR) models to exploit spatial correlations as well as also introduce a 3-step approach to improve forecast accuracy.