Controlling airline seat allocations with neural networks

Presents a neural network that is intended to support airline marketing specialists in controlling seat allocations on flight departures. The focus of the investigation is the prediction of overbooking rates in order to avoid a situation where an aircraft departs with empty seats when passengers who have booked seat do not participate in the flight. The neural network proposed to solve the problem is an extension of the forward-only counterpropagation model. The network learns to approximate the mapping between the input data (the number of booked seats for each reservation class at distinct time periods prior to departure) and the desired output (the number of no-shows). The trained network is then used to make the predictions for the future. The feasibility of this approach is demonstrated by an efficient implementation