A Hybrid Neuro-Fuzzy Model for Stock Market Time-Series Prediction

摘要

In this paper we propose a hybrid five-layer neuro-fuzzy model and a corresponding learning algorithm with application in stock market time-series prediction tasks. The key difference between classical ANFIS architecture and the proposed model is in the fourth layer - multidimensional Gaussian functions are used instead of polynomials in order to achieve better computational performance and representational abilities in processing highly nonlinear volatile data. The experimental results have shown the clear advantages of the described model and its learning.