Investigating optimal feature selection method to improve the performance of Amharic text document classification

Abstract

Feature selection is one of the famous solutions to reduce high dimensionality problem of text categorisation. In text categorisation, selection of good features (terms) plays a crucial role in improving accuracy, effectiveness and computational efficiency. Due to the nature of the language, Amharic documents suffered from high dimensionality feature space that degrades the performance of the classifier and increases the computational cost. This paper investigates optimal feature selection methods for Amharic Text Document Categorisation among various feature selection techniques such as Term Frequency*Inverse Document Frequency (tf*idf), Information Gain (IG), Mutual Information (MI), Chi-Square (-X²), and Term Strength (TS) using Support Vector Machine (SVM) classifiers. Experimentations carried out based on the collected datasets showed that X² and IG method performed consistently well on Amharic document Texts among other methods. Using both methods, the SVM classifier showed a significant improvement of the classification accuracy and computational efficiency.

Keywords: Feature selection, Amharic, SVM, Classification