Generating rules by inductive machine learning

Abstract

Knowledge is a basic component of expert system. But knowledge acquisition for business problems like stock investment is difficult because the knowledge is unstructured and time dependent. Therefore knowledge acquisition via domain expert is not suitable. In this research, based on the Quinlan``s ID3, the knowledge acquisition by machine learning was adopted. This machine learning generates knowledge from a large data base of examples by inductive learning. The knowledge is rule based type and is represented by tree structure. This learning system selects the most discriminatory attribute first and put the attribute on root node of tree, and selects next most discriminatory attribute as the root of lower level subtree. This process continues until all discriminating attributes are included in the rule base. The measure of discriminatory power of an attribute is entropy and the strength of a rule is measured by certainly factor. The system was constructed in supermicro computer and exploratory knowledge on stock investment is generated. This system is general enough to apply to any domain as far as the structure of examples is the combination of attributes and values.