A Comparative Study of Inductive Learning Algorithms

摘要

This paper reviews the five main categories of inductive learning algorithms: ID3, AQ, RULES, IT and SAFARI. The algorithms are compared using a number of salient features. These features include generalisation, sensitivity to producing the same set of rules, robustness to noise, irrelevant value problem, missing branches problem, bias problem, simplicity, and nature of induction process. Speed of inferencing and time complexity of extracting rules using various inductive learning algorithms are discussed. Good classification accuracy is vital for a number of applications, the algorithms are tested on IRIS data set for classification accuracy comparison. The size of the set of rules is important to fast inferencing, a comparison of the size of the set of rules obtained by the five main types of algorithms on IRIS data set is performed.