Intelligent Systems and Web Agents for Financial Forecasting

Abstract

Recently, artificial intelligence (AI) is popularly applied to the problems of finance such as stock market prediction, bankruptcy prediction, and corporate bond rating. In particular, several studies on stock market prediction using AI techniques have been executed during the past decade. The reason is that it can model nonlinear relationships among financial variables. Some of them, however, did not provide outstanding prediction accuracy partly because of the tremendous noise and complex dimensionality in stock market data. The noise and complex dimensionality can be reduced through proper preprocessing. In addition, sometimes the amount of data is so large that the learning of patterns may not work well. Another reason of this inconsistent and unpredictable performance of prior research may be problems associated with the ad hoc nature for designing AI techniques. First, this study proposes feature transformation approaches using domain knowledge and genetic algorithms (GAs) to mitigate the limitations of prior research. Proposed approaches produce significantly better performance than conventional approaches. Second, this study proposes simultaneous optimization approaches using the GA for AI techniques. Prior research has used global or local search algorithms to optimize architectural factors in AI techniques, but they included only a part of these factors in their consideration. If these factors are considered separately, global optimization is achieved in part, but may lead to locally optimized solutions as a whole. However, if these factors are simultaneously considered, the performance may be enhanced because it will cause the optimization of all factors in a synergistic way then it may lead global optimization as a whole. Experimental results show that the GA approaches to simultaneous optimization of AI techniques are viable alternative approaches for stock market prediction. Third, this study proposes the rough set approaches and the GA as methods...