A technical analysis expert system with knowledge refinement mechanism

A refinement system is given which tries to refine a rule base, using task and domain strategies. In particular, the task strategy has domain-free rule primitives which are just relevant to generic tasks (signal processing or pattern recognition) and tries to move the initial rule space into a new better one, testing the applicability of new rule candidates in the new rule space with real data. An experiment on real Japanese stock-price data shows that a refined rule base almost has the same performance as one with top-level technical analysts in the Japanese stock market. This result implies that task and domain strategies would be essential refinement operators to move a rule space into a better one