The Role of Interval Initialization in a GBML System with Rule Representation and Adaptive Discrete Intervals

Abstract

This paper examines some initialization methods for a genetic based machine learning (GBML) rule representation [2] which works with adaptive discretization intervals. The methods studied apply different degrees of uniformness to the initial intervals of the population. The tests done show that except the test problems with more attributes, the differences between the tested methods accuracies are not significant. This proves that we only have to be aware of it in a limited kind of problems.