Selection and Statistical validation of features and prototypes

摘要

Features and protypes selection are two major problems in data mining,especially for machine learning algorithms.The goal of both selectirons is to reduce storage complexity,and thus computational costs,without sacrificing accuracy.In this article,we present two incremental algorithms using geometrical neighborhood graphs and a new statistical test to select,step by step,relevant features and prototypes for suppervised learning problems.The feature selection selection procedure we present could be applied before any machine learning algorithm is used.