Improving Performance of a Binary Classifier by Training Set Selection

In the paper a method of training set selection, in case of low data availability, is proposed and experimentally evaluated with the use of k-NN and neural classifiers. Application of proposed approach visibly improves the results compared to the case of training without postulated enhancements. Moreover, a new measure of distance between events in the pattern space is proposed and tested with k-NN model. Numerical results are very promising and outperform the reference literature results of k-NN classifiers built with other distance measures.