Forecast of Aircraft Retirement Probability using Neural Networks

Abstract

In this paper, we present a novel approach for the prediction of retirement curves, which summarize the survival / retirement probabilities of aircraft. Retirement curves are used to predict the future aircraft fleet composition, as an element of air traffic and emissions forecasts. Furthermore, retirement curves are a tool for aircraft manufacturers and leasing companies to estimate the need for replacement aircraft as part of global aircraft demand. We have applied a methodology involving neural networks, previously being used in the area of predictive maintenance. Transferring this method of data analysis to a new research field goes beyond previously applied methodologies, as neural networks are known for finding connections in data that cannot be explicated with other methodologies. In the context of retirement curves, neural networks can help to explain the influence of economic factors or general market development on aircraft retirement. We implement a neural network to calculate survival probabilities of aircraft and estimate the impact of economic framework data on survival probabilities. The neural network uses the global passenger aircraft fleet as training data base and predicts survival probabilities which then are arranged into retirement curves. We then compare the results to other retirement curve prediction methods. Finally, we take an outlook on possible enhancements of this method and amplification options to improve the quality of the forecast.