Inductive Learning for Continuous Classes and the Effect of RULES Family

Abstract

Abstract-Due to the digital world explosion and its appearance in everyday life, predicting numeric actions became necessary in machine learning. However, due to growth of interest in understanding how problems can be solved, simple prediction algorithms are more helpful than the difficult statistical approaches. Covering algorithm can be used to accomplish difficult problems using simple rules or trees. One family called RULES was found to be very interesting with appealing properties. It is one of the most flexible and simplest families with high learning rate. Nevertheless, even though RULES is actively improving but it is surprisingly neglected, especially with numerical datasets. Thus, the purpose of this paper is to extend the literature and investigate the problems of continuous classes in RULES and other inductive learning families. A theoretical analysis is conducted to show the effect of numerical actions and how it is still an open research area. An empirical evaluation is also provided to prove how RULES family can be used as the base of further improvement. Accordingly, this paper can be used as a reference by researchers to know what research area is still not covered and need further refinement in inductive learning.