Revenue Management – Artificial Neural Network approach to overbooking problem

Abstract

The aim of this dissertation is to propose an artificial neural network model that would most accurately predict the overbooking for 3 booking classes of LOT Polish Airlines’ flight LO381. It is proved that accuracy of the model is superior to statistical tools used widely nowadays and the application of neural network for solving the overbooking prediction is highly motivated. What is more, it is demonstrated that proposed model is even more accurate than the model given by Freisleben and Gleichmann in their article ‘Controlling Airline Seat Allocations with Neural Networks’ from IEEE Transactions on Neural Networks from 1993. In the first part Revenue Management and its history is introduced. This part of the paper explains also what the Revenue Management consists of and explains in detail the overbooking and seat inventory control. The second part concerns the Artificial Neural Networks. In this part, the background of the neural science as well as its history is described shortly. The types of Artificial Neural Networks and several learning rules are characterized. Finally, the detailed description of Kohonen’s Self-Organizing Map is presented. The last part of the dissertation includes description of the Artificial Neural Network application for solving the overbooking problem. The data set used to build the network and its topology is described. Moreover, the error and prediction analyses are also characterized in this section.