The construction of computerized classification systems using machine learning algorithms: An overview

摘要

Although there are several microcomputer programs that are concerned with the automatic induction or machine learning of classification rules, the techniques seem to have been largely ignored in favor of conventional parametric procedures. Various machine learning algorithms are discussed, and the available evidence comparing their performance with linear discriminant function (LDF) analysis is reviewed. Although far more research is required, there are indications that machine learning algorithms are capable of giving better results than LDF. Similar algorithms known as automatic interaction detectors, originally employed in the field of survey analysis, are also examined. It is concluded that a computer program representing a synthesis of machine learning techniques, conventional statistical methods, and automatic interaction detectors should be developed.