CFS Based Feature Subset Selection for Enhancing Classification of Similar Looking Food Grains- A Filter Approach

Feature selection plays an important role in machine learning. The best selection of significant feature set results in better performance of classifiers. In this work feature selection using filter approach for classification of 10 different similar looking food grains has been carried out. As part of the work, firstly texture features of food grain images are extracted using color based GLCM. As part of feature subset selection, the Correlation-based Feature Selection (CFS) filter approach has been applied with 5 different search methods. The evaluation of feature selected is done using 6 different classifiers. Result shows the feature set selected by the CFS filter have indeed enhanced the classification performance of all the 6 classifiers when compared to their performance with the original feature set.

I. Introduction

Because of computerization, huge amount of data is getting in accumulated. Volume of data both in terms of dimensionality and number of records has become one of the challenges for machine learning and data mining tasks has increased. To overcome this problem of high dimensionality, dimensionality reduction techniques play a vital role in selecting the reduced set of relevant features which not only improves the classification performance but also reduces computation time. The main objective of feature reduction is to find minimum set of attributes such that the resulting probability distribution is obtained using all attributes [1] .