A decision support tool for tuning parameters in a machine learning algorithm
Margot Postema, Tim Menzies, Xindong Wu · 1997
Many machine learning algorithms require parameter tuning in order to adapt them to the particulars of a training set. This tuning task can be an expert task in its own right. Based on our observations of an expert tuning the HCV (Version 2.0) rule induction algorithm, we have built a simple decision support system (DSS) to automate tuning for that algorithm. We found that HCV plus DSS produced equal or better rule sets than standard HCV or C4.5. A surprising result from this study is that a very simple approach to tuning worked very well. We speculate that other research on machine learner tuning have indulged in complex solutions before experimenting adequately with simpler alternatives. Keywords: Decision support systems, machine learning, discretization, fine tuning 1 Introduction The dream of machine learning is that we can automatically build specifications from data without requiring tedious and time consuming human involvement. To date, this dream has not been realised. A repe...