Improving Forecasting Accuracy by Neural Network Applying to the Airlines Cargo Data in the Case of Daily Data

Abstract

In recent years, severe competition is executed on getting air cargos. The forecast of the number of taking-off and landing is expanding. Strict marketing is required in such fields. Forecasting the trend of air cargo is an essential item to be investigated in airlines. In order to make forecast for time series, the method of using linear model is often used. Forecasting using neural network is also developed. Reviewing past researches, there are many researches made on this. There is many room to improve in neural network, therefore we make focus on them. We use time series data, and in order to make forecast, a new coming data should be handled and the parameter should be estimated based upon its data. This is a so-called on-line parameter estimation. In this paper, neural network is applied and Multilayer perceptron Algorithm is newly developed. The method is applied to the Airlines Cargo Data in the case of Daily data. When there is a big change of the data, the neural networks cannot learn the past data properly, therefore we have devised a new method to cope with this. Repeating the data into plural section, smooth change is established and we could make a neural network learn more smoothly. The result is compared with the method of ARIMA model. The forecasting results are measured by the Forecasting Accuracy Ratio which is the measure of the normalized residual part of the forecasting error. Good results were obtained. The new method shows that it is useful for the time series that has various trend characteristics and has rather strong seasonal trend. The effectiveness of this method should be examined in various cases.

Keywords

• forecasting
• neural network
• time series analysis
• ARIMA model