In machine learning, the learner is presented with a set of instances and its associated target values to learn. The learner makes some assumptions and builds a model, also called as classifier which is a hypothesis constructed by the learning algorithm. Out of the N input variables in the training instances not all are relevant to build the classifier model. Out of N, only k variables are relevant to build the hypothesis with respect to target output where k << N (k is very leass as compared to N). Often, the task of finding the relavnt k variables is not straight-forward.

Stracuzzi and Utgoff (2004) propose a novel approach to deal with this problem. In their paper they have given an alogorithm to use randomized variable selection technique to find the 'k' relevant variables. I have implemented the algorithm in Java, obtained the results and matched them with those given by the authors of this paper. The results of my implementation are given on this site.