An application of fuzzy time series to improve ISE forecasting

Abstract

The problem of fuzzy time series forecasting plays an important role in many scientific areas such as statistics and neural networks. While forecasting fuzzy time series, most of forecasting applications use the same length of intervals. The determination of length of intervals is significant and critical in fuzzy time series forecasting. The usage of convenient performance measure may also have an important affect for forecasting studies. MSE (Mean squared error) as a performance measure is widely used in many studies. The aim of this paper is to improve fuzzy time series forecasting by using different length of intervals with neural networks according to various performance measures. For this reason, we take ISE (Istanbul stock exchange) national-100 index as a large data set for forecasting. We use various performance measures such as MSE, RMSE (Root mean squared error), MAE (Mean absolute error) and MAPE (Mean absolute percentage error) to compare forecasting performances with different length of intervals. The empirical results show that the most convenient length of intervals can be chosen as 300 by comparing overall performance of MSE, RMSE, MAE and MAPE by using neural networks.