Machine learning feature selection of financial data

Abstract

Global financial crisis has been occurred frequently in these days also na- tions and companies pay attention to predict bankruptcy. In this thesis, we discuss feature selection method to extract feature group that causes main factors of making bankruptcy. We describe stepwise method and principal component analysis method briefl y and compare it to construct prediction model. In addition, we try to analyze their performance and statistical mea- surement which method is the most efficient to raw data. We deal with data set of experiments which consist of 515 companies' financial statement in 1997 to build the model by using support vector machine.