Learning compositional decision rules using the KEX algorithm

Abstract

If-then rules belong to the most popular formalism used to represent knowledge either obtained from human experts as in the case of expert systems or learned from data as in the case of machine learning and data mining. The most commonly used approach to learning decision rules is the set-covering approach, also called separate and conquer. The other way to create decision rules is the compositional approach. The work reported in this paper fits into the latter approach. We will describe the KEX algorithm, its implementation within the LISp-Miner system, and results of empirical comparison of KEX with some other rule-learning algorithms implemented in the Weka system.