Exception Included, Ordered Rule Induction from the Set of Exemplars (ExIORISE)

Abstract

An expert system is one which uses collection of data comprising the knowledge to offer guidance or make inferences. Its work in most cases can be seen as classification which is basically the task of assigning objects to different categories or classes, determined by the properties of those objects. Numerous research works have been and being done to develop efficient knowledge acquisition techniques for expert systems. Some state-of-the-art algorithms are great performers but need extensive learning whereas older rule / decision- tree based algorithms perform pretty well with small data sets. Moreover, co-existence of learners of different levels of expertise and accuracy is believed to be encouraged to achieve a cumulative intelligence just like the human beings have. RISE is one such algorithm that infuses both instance-based learning and rule induction. It proved to be quite efficient handling binary and multi-class classification problems for small data sets in terms of accuracy and cost as well. In this work, features like exclusion of inefficient rules, inclusion of exceptions in the rule set and ordering of the rules using weights beforehand are integrated with the classical RISE algorithm to develop a more efficient classifier system named as ExIORISE. Empirical study shows that ExIORISE outperforms RISE, C4.5 and CN2 significantly.