A Hybrid KNN-Ant Colony Optimization Algorithm for Prototype Selection

The condensing KNN is the application of the K-Nearest Neighbors classifier with a condensed training set, which is a consistent subset calculated from the initial training set. In this work we present a novel algorithm, Ant-KNN, which allows improving the performance of the standard KNN classifier by a method based on ant colonies optimization. The results obtained through tests conducted on five benchmarks from UCI Machine Learning Repository demonstrate the improvement obtained by our algorithm in comparison with other condensing KNN algorithms.