Rule Quality Measures Improve the Accuracy of Rule Induction: An Experimental Approach

Abstract

Rule quality measures can help to determine when to stop ge- neralization or specification of rules in a rule induction system. Rule quality measures can also help to resolve conflicts among rules in a rule classification system. We enlarge our previous set of statistical and empirical rule quality formulas which we tested earlier on a number of standard machine learning data sets. We describe this new set of formulas, performing extensive tests which also go beyond our earlier tests, to compare these formulas. We also specify how to generate formula-behavior ru- les from our experimental results, which show the relationships between a formula’s performance and the characteristics of a dataset. Formula- behavior rules can be combined into formula-selection rules which can select a rule quality formula before rule induction takes place. We report the experimental results showing the effects of formula-selection on the predictive performance of a rule induction system.