Applying artificial neural networks to managing stock subject to erratic demand

A good many products in an inventory list of hundreds or even thousands of items are subject to very sporadic demand. Such items, when ordered, are usually ordered in quantities of more than one per order. Representing demand over the stocking period on the basis of continuous distribution such as normal or exponential (methods) can lead to major error. Many analytical studies attempting to model both the time between orders and quantities required have already been carried out, and have come up with complex processes (one such example would be the Poisson Spasmodic Process). These are difficult to implement and provide solutions which are admittedly improvements on those that preceded them, but which are nevertheless seen as being far from ideal. This work proposes applying Artificial Intelligence techniques such as Artificial Neural Networks to the modelling process for time series describing products subject to erratic demand. If more realistic forecasts can be obtained, inventory management can better correspond with the real supply requirements for these products.