Predicate ranking algorithms and their application in an inductive logic programming system

Abstract

Inductive logic programming (ILP) is a form of machine learning that induces rules from data using the language and syntax of logic programming. A rule construction algorithm forms rules that summarize data sets. These rules can be used in a large spectrum of data mining activities. In ILP, the rules are constructed with a target predicate as the consequent, or head, of the rule, and with high-ranking literals forming the antecedent, or body, of the rule. The predicate rankings are obtained by applying predicate ranking algorithms to a domain (background) knowledge base. We present three new predicate ranking algorithms for the inductive logic programming system, INDED (pronounced "indeed"). The algorithms use a grouping technique employing basic set theoretic operations to generate the rankings. We also present results of applying the ranking algorithms to several problem domains, some of which are universal like the classical genealogy problem and others, not so common. In particular, diagnosis is the main thread of many of our experiments. Here, although our experimentation relates to medical diagnosis in diabetes and Lyme disease, many of the same techniques and methodologies can be applied to other forms of diagnosis including system failure, sensor detection, and trouble-shooting.