IMPROVING FORECASTING ACCURACY FOR TIME SERIES DATA USING FUZZY TECHNIQUES AND WAVELET TRANSFORM

Abstract

This study focuses on improving the accuracy of stock market forecasting for the Saudi Arabia stock exchange (Tadawul) by employing advanced modeling techniques and adaptive learning approaches. The study utilizes the maximum overlapping discrete wavelet transform (MODWT) in conjunction with various mathematical functions to analyze daily stock price indices data from October 2011 to December 2019. Input variables, including oil price and repo rate, are carefully selected based on correlation analysis, multiple regression, and the Engle and Granger Causality test. The proposed models, such as MODWT-LA8-ANFIS, MODWT-LA8-FS.HGD, MODWT-LA8-HyFIS, and MODWT-LA8-FIR.DM, demonstrate superior forecasting performance compared to traditional methods like ARIMA, ANFIS, FS.HGD, HyFIS, and FIR.DM. The performance evaluation of the proposed model involves various statistical measures, including mean error (ME), root mean square error (RMSE), mean absolute error (MAE), and mean percentage error (MPE). The results highlight the effectiveness of these models in decomposing stock market patterns and accurately predicting stock market price volatility. This research contributes to the field of stock market forecasting and offers valuable insights for investors and financial analysts operating in the Saudi Arabia stock exchange.