A real-time data mining with genetic algorithm

초록

The purpose of this paper is to present a new mining algorithm for mining a real-time data. This study uses both the genetic algorithm and the rule induction using decision tree to predict stock market index. Input variables used in this study are slow% D, ROC, William% R, CCI. Output variable was also used as KOSPI index. Research procedures are conducted in the following orders: Data entry after data division, agents (rule extraction), performance tests, agent mutation, prediction rate tests. As a result of the analysis, prediction rate in each generation improves the decision making ability of the agents by acquiring traits to adapt to the environment. This shows that the ability for prediction performance increases by obtaining a dominant genes. Evolution leads to the better prediction accuracy. This presents that decision-making capabilities of the agent is improved by acquiring a genetic trait to adapt to the environment. In other words, agents continually make a genetic mutation in order to adapt to environmental changes, and the new child agents having better predictive genes survive. In this paper, we compared the proposed algorithm with other methodologies to verify the effectiveness of the model. We performed experiments that compared the proposed algorithm with rule induction methods, neural network, discriminant analysis and logistic regression analysis. All experiments were performed in the same situation. The proposed method showed the best predictive performance. As the proposed algorithm is more effective than other conventional mining techniques, this study will have a good chance that can effectively apply a real-time data in real business environment. © Serials Publications Pvt. Ltd.

키워드