Using Genetic Algorithms to Optimize Neural Network and Applying It to Forecast Electronics Stocks Index

• Main Authors: Chen, Tien-Shang, 陳天賞
• Other Authors: Chiu, Deng-Yiv
• Format: Others
• Language: zh-TW
• Published: 2003
• Online Access: http://ndltd.ncl.edu.tw/handle/47455171232990877892

碩士 === 中華大學 === 資訊管理學系 === 91 === In the recent years, more researches used artificial neural network (ANN) to forecast stocks price. In these researches, the performance of prediction models was always satisfactory. But, these researches lacked a criterion rule to choose input variables and optimal network architecture. Most of them used trial-and-error to find a better network architecture in some range. Therefore, in this thesis, we use genetic algorithms (GA) to optimize neural network and use the gradient steepest descent method and BPN/Cauchy machine as the learning algorithm to train network weights individually. We hope to improve ability of network prediction model through our proposed methods. At last, we apply our proposed methods to forecast electronics stocks index. In the results of experiments, we conclude five conclusions as follows. First, the parameters and architecture of neural network must be chosen by network. Secondly, there is not only one best component architecture of prediction model. There may be various different component architectures of prediction models. Thirdly, the gradient steepest descent method and BPN/Cauchy machine learning algorithms have some advantages and disadvantages individually. Which learning algorithm to choose must be dependent on the problem to solve. Fourthly, the performance of prediction model is satisfactory after optimizing neural network. At last, in the results of experiment, our proposed method is feasible for optimizing neural network.