Dynamic Recursive-Based Optimal Learning Fuzzy Systems Design－Constructing Financial Time-series Prediction System

• Main Authors: Hsiang-Chai Chou, 周祥在
• Other Authors: Hsuan-Ming Feng
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/19935391766580764909

碩士 === 國立金門技術學院 === 電資研究所 === 97 === The tremendous sudden variation and complex nonlinear dimensionality of the stock prices is a big challenging task in stock markets predicting and trading problems. In this thesis, the dynamic recursive-based optimal learning algorithm is designed to build financial time-series fuzzy prediction systems. The fuzzy system with suitable human-beings decision rules is determined to identify behavior of the discussed time-series stock data. Containing with the adapt ability of the fuzzy inference system, the proposed fuzzy prediction systems can reduce the affect in some of non-quantifiable political, social noise and interference. The objective of this stock prediction system is to reduce investment risk, and enable investors to reap the maximum profit. The dynamic clustering method is first applied to configure the initial architecture of the fuzzy prediction system. Selected fuzzy-rules number is equal to the number of cluster centers and cluster center results will be assigned to locate the initial center position of the membership function. The particle swarm optimizationm (PSO) and recursive least square (RLS) methods are integrated to tune fuzzy system parameters for approximating toward the trade feature of the collected Taiwan's weighted index. Experiment results compared with other machine learning schemes for the prediction of the stock prices and tracking trend are illustrated to demonstrate our proposed system having better sell and buy winning stratagem.