Demand Models For Supermarket Demand Forecasting

Model-based approaches remain an important option for modelling customer demand. While this approach allows to analyse demand using a model based on theoretical arguments, each choice of model is associated with specific  assumptions under which this model is valid. Customer demand in retail is typically modelled as a Poison-type process, in particular using a negative binomial distribution. Poisson-type processes are associated with an exponential inter-arrival time that describes the probability distribution between subsequent events. Using a public dataset from a large  supermarket, the analysis of the data shows that while the general assumption of a Poisson process is reasonable, the purchasing behaviour strongly depends on the type of product. Additionally, customers in this supermarket show astrong preference for a weekly shopping trip.