A neuro-fuzzy based method for TAIEX forecasting

Time series prediction can be widely applied to a variety of fields. Recently, a lot of artificial intelligence (Al) techniques have been exploited in the task of time series prediction. Compared to statistical methods, Al techniques are easier to use for real world data, and their performance can be better. In this paper, we propose a neuro-fuzzy based system for time series prediction. The neuro-fuzzy based system can generate superior performance through the relationship among different features. By partitioning the training data into clusters, fuzzy IF-THEN rules are extracted to form a fuzzy rule-base. Then, a fuzzy network is constructed accordingly and parameters are refined to increase the precision of the fuzzy rule-base by applying a hybrid learning algorithm which combines a recursive singular value decomposition-based least squares estimator and the gradient descent method. We demonstrate the effectiveness of the proposed system by applying it to do prediction for TAIEX stock indices. The experimental results conclude the superiority of the proposed system over other existing systems.