Simultaneously Removing Noise and Selecting Relevant Features for High Dimensional Noisy Data

Abstract

The classification for the noisy training data in high dimension suffers from concurrent negative effects by noise and irrelevant/redundant features. Noise disrupts the training data and irrelevant/redundant features prevent the classifier from picking relevant features in building the model. Therefore they may reduce classification accuracy. This paper introduces a novel approach to improve the quality of training data sets with noisy dependent variable and high dimensionality by simultaneously removing noisy instances and selecting relevant features for classification. Our approach relies on two genetic algorithms, one for noise detection and the other for feature selection, and allows them to exchange their results periodically at certain generation intervals. Prototype selection is used to improve the performance along with the genetic algorithm in the noise detection method. This paper shows that our approach enhances the quality of noisy training data sets with high dimension and substantially increases the classification accuracy.