Features selection, data mining and finacial risk classification: a comparative study

• Author: Salim Lahmiri
• Published date: 01 October 2016
• Date: 01 October 2016
• DOI: http://doi.org/10.1002/isaf.1395

RESEARCH ARTICLE

Features selection, data mining and finacial risk classification: a

comparative study

Salim Lahmiri

ESCA School of Management, Casablanca,

Morocco

Correspondence

Salim Lahmiri, ESCA School of Management,

Casablanca, Morocco.

Email: slahmiri@esca.ma

Summary

The aim of this paper is to compare several predictive models that combine features selection

techniques with data mining classifiers in the context of credit risk assessment in terms of accu-

racy, sensitivity and specificity statistics. The t‐statistic, Battacharrayia statistic, the area

between the receiver operating characteristic, Wilcoxon statistic, relative entropy, and genetic

algorithms were used for the features selection task. The selected features are used to train

the support vector machine (SVM) classifier, backpropagation neural network, radial basis func-

tion neural network, linear discriminant analysis and naive Bayes classifier. Results from three

datasets using a 10‐fold cross‐validation technique showed that the SVM provides the best accu-

racy under all features selections techniques adopted in the study for all three datasets. There-

fore, the SVM is an attractive classifier to be used in real applications for bankruptcy

prediction in corporate finance and financial risk management in financial institutions. In addition,

we found that our best results are superior to earlier studies on the same datasets.

KEYWORDS

classification, credit risk, data mining, features selection

1|INTRODUCTION

In corporate finance, financial risk prediction is important for financial

decision‐making and profitability of financial institutions. Therefore,

several predictive models based on data mining techniques have been

developed to accurately classify bankrupted and non‐bankrupted com-

panies. The problem of financial risk prediction is important for finan-

cial institutions and investors for better risk control (Cochran et al.,

2006; Sanz and Ayca; 2006; Pindado et al., 2008; Abdou and Poiton,

2011; Platt and Platt, 2012; Savona and Vezzoli, 2012; Çelik, 2013;

Bastos and Pindado, 2013; Evans and Borders, 2014; Mendes et al.,

2014; Ciampi, 2015). Therefore, the topic has received much attention

in corporate finance and risk management literature. Indeed, several

data mining techniques have been used for the prediction of financial

risk, including the AdaBoost algorithm (Alfaro et al., 2008), fuzzy

support vector machine (SVM) (Chaudhuria and De, 2011), SVM (Song

et al., 2010; Horta and Camanho, 2013), backpropagation neural net-

work (BPNN) (Trinkle and Baldwin, 2007; Peat and Jones, 2012; Lee

and Choi, 2013), radial basis function neural networks (RBFNNs)

(Divsalar et al., 2011), decision tree algorithms (Delen et al., 2013),

ensemble of SVMs (Sun and Li, 2012), linear programming (Divsalar

et al., 2011), if–then rules (Davalos et al., 2014), genetic algorithms

(GAs) (Gordini, 2014), ensemble systems (Sun, 2012; Figini et al.,

2016), case‐based reasoning system (Li et al., 2013), probabilistic neu-

ral network (Pendharkar, 2011) and fuzzy neural approach (Quek et al.,

2009),

Recently, several studies have adopted a particular feature selec-

tion technique to improve the classifier accuracy in predicting financial

risk. Indeed, the goal of feature selection is to identify the most infor-

mative features used as patterns in the classification task and remove

redundant ones. In this regard, selected features are expected to

improve classification/prediction results and help in reducing the pro-

cessing time of the classifier. For instance, these studies used t‐statistic

(Ravisankar et al., 2011), principal component analysis (Chen, 2012;

Lin, 2012), partial least square (Yang et al., 2011; Serrano‐Cinca and

Gutiérrez‐Nieto, 2013), stepwise discriminant analysis (Li and Sun,

2011), information gain (Wang et al., 2014), GAs (Oreski and Oreski,

2014), decision trees (Cho et al., 2010), classification tree method

(Brezigar‐Masten and Masten, 2012) and logistic regression (Lahmiri

and Gagnon, 2015) to perform the features selection task.

However, it is difficult to select an adequate combination of fea-

tures selection technique and classifier for a given dataset, which is

an important task in bankruptcy prediction. Indeed, the performance

of the classifier depends on the features selection techniques and

DOI 10.1002/isaf.1395

Intell Sys Acc Fin Mgmt 2016; 23: 265–275Copyright © 2016 John Wiley & Sons, Ltd.wileyonlinelibrary.com/journal/isaf265