An algorithm which learns multiple covers via integer linear programming. Part I: the CLILP2 algorithm

Abstract

Presents an inductive machine learning algorithm called CLILP2 that learns multiple covers for a concept from positive and negative examples. Although inductive learning is an error‐prone process, multiple meaning interpretation of the examples is utilized by CLILP2 to compensate for the narrowness of induction. The algorithm is tested on data sets representing three different domains. Analyses the complexity of the algorithm and compares the results with those obtained by others. Employs measures of specificity, sensitivity, and predictive accuracy which are not usually used in presenting machine learning results, and shows that they evaluate better the “correctness” of the learned concepts. The study is published in two parts: I – the CLILP2 algorithm; II – experimental results and conclusions.