An Empirical Analysis of Data Reduction Techniques for k-NN Classification

Abstract

This study explores Data Reduction Techniques (DRTs) in the realm of lazy classification algorithms like k-NN, focusing on Prototype Selection (PS) and Prototype Generation (PG) methods. The research provides an in-depth examination of these methodologies, categorizing DRTs into two primary categories: PS and PG, and further dividing them into three sub-categories: condensation methods, edition methods, and hybrid methods. An experimental study compares a total of 20 new and state-of-the-art DRTs across 20 datasets. The objective is to draw performance conclusions within both the primary and sub-categories, offering valuable insights into how these techniques enhance the effectiveness and robustness of the k-NN classifier. The paper provides a comprehensive overview of DRTs, clarifying their strategies and relative performances.