Discovering Knowledge Nuggets with a Genetic Algorithm

摘要: Measuring the quality of a prediction rule is difficult task, which can involve several criteria. The majority induction literature focuses on discovering accurate, comprehensible rules. In this chapter we also take these two criteria into account, but go beyond them in sense that aim at rules are interesting (surprising) for user. Hence, search guided by rule-evaluation function considers both degree predictive accuracy and interestingness candidate performed versions genetic algorithm (GA) specifically designed to discovery - or “knowledge nuggets.” addresses dependence modeling task (sometimes called “generalized induction”), where different predict goal attributes. This be regarded as generalization very well known classification all same attribute. compares results GA with simpler, greedy discover