A comparison of forecasting methods for rotable spare parts in a Thai low-cost airline

Abstract

This research intents to compare and evaluate 3 different methods for forecasting irregular demand of aircraft spare parts by focusing on Rotable part group of Company A which is the leading low-cost airline in Thailand and operate by Airbus A320 model. Based on the demand classification, it is observed that most of the spare part demands presented an intermittent demand pattern. The forecasting methods Croston’s, Holt’s Linear and Multi-Layer Perceptron trained with Backpropagation of Neural Network Model are applied on the historical data of 36 Rotable spare parts by dividing into 2 sets which are initialization set period 60 months (January 2014 – December 2018) and test set for period 12 months (January 2019 – December 2019). The forecasting accuracy will be made by using ME, MAE and A-MAPE to evaluate the performance of the method over the test set. The results show that Multi-Layer Perceptron trained with Backpropagation of Neural Network Model outperform than other 2 methods in all forecasting accuracy measurement and the performance 81% accurate forecast when comparing with the actual observation over the test set data in 2019 for 12 months period. Therefore, from this research it concludes that Multi-Layer Perceptron trained with Backpropagation of Neural Network Model is superior alternative model over the traditional time series models and recommend to be a forecasting tool for Company A use to forecast Rotable spare part in intermittent demand pattern.