Evolutionary Fuzzy Stock Prediction System Design and Its Application to the Taiwan Stock Index

The study develops fuzzy stock prediction system that integrates the novel computer technologies of stepwise regression analysis (SRA), auto-clustering analysis, recursive least-squares (RLS) and particle swarm optimization (PSO) learning schemes. The SRA methodology serves as a data filtering channels to select two primary technical indexes from the training dataset. The selected items are then assigned as input variables of the fuzzy prediction system to simplify the modeling architecture. An efficient evolutionary clustering algorithm can then determine the available centers positions of the membership functions. It is proposed to exploit appropriate behaviors of the identified stock dataset. In addition, the initial architecture of fuzzy prediction system is represented with the trained clusters information. The proposed evolutionary-based learning scheme with its evolutionary hybrid particle swarm optimization (PSO) and recursive least-squares (RLS) technologies can extract the appropriate parameters for the fuzzy stock prediction system. The proposed fuzzy stock prediction systems are implemented as necessary data acquirement, feature description and time serious stock prices and trends forecasting stages in the real financial environment. The objective of this study is that it is not only capable of automatically initializing and creating the appropriate fuzzy architecture, but it can also develop the stock model to accurately simulate the actual trading of the Taiwan stock indexes (TAIEX). This study develops various stock predictions examples for daily and weekly approximations for the training and testing phases based on historical TAIEX data. A comparison with other learning methods shows that the proposed approach offers improved forecasting accuracy. The generated fuzzy stock model can help traders achieve the greatest rewards in real TAIEX trading applications.