Performance assessment of ensemble learning systems in financial data classification

• Published date: 01 January 2020
• Author: Salim Lahmiri,Stelios Bekiros,Anastasia Giakoumelou,Frank Bezzina
• DOI: http://doi.org/10.1002/isaf.1460
• Date: 01 January 2020

RESEARCH ARTICLE

Performance assessment of ensemble learning systems in

financial data classification

Salim Lahmiri

1

| Stelios Bekiros

2,3

| Anastasia Giakoumelou

4

| Frank Bezzina

5

1

Department of Supply Chain & Business

Technology Management, John Molson School

of Business, Concordia University, Montreal,

Canada

2

Department of Economics, European

University Institute, Florence, Italy

3

Rimini Centre for Economic Analysis, Wilfrid

Laurier University, Waterloo, Canada

4

Department of Law and Management, School

of Economics, University of Rome Tor Vergata,

Rome, Italy

5

Faculty of Economics, Management, and

Accountancy, University of Malta, Msida,

Malta

Correspondence

Salim Lahmiri, Department of Supply Chain &

Business Technology Management, John

Molson School of Business, Concordia

University. Montreal, Canada.

Email: salim.lahmiri@concordia.ca

Summary

Financial data classification plays an important role in investment and banking indus-

try with the purpose to control default risk, improve cash and select the best cus-

tomers. Ensemble learning and classification systems are becoming gradually more

applied to classify financial data where outputs from different classification systems

are combined. The objective of this research is to assess the relative performance of

existing state-of-the-art ensemble learning and classification systems with applica-

tions to corporate bankruptcy prediction and credit scoring. The considered ensem-

ble systems include AdaBoost, LogitBoost, RUSBoost, subspace, and bagging

ensemble system. The experimental results from three datasets: one is composed of

quantitative attributes, one encompasses qualitative data, and another one combines

both quantitative and qualitative attributes. By using ten-fold cross-validation

method, the experimental results show that AdaBoost is effective in terms of low

classification error, limited complexity, and short time processing of the data. In addi-

tion, the experimental results show that ensemble classification systems outperform

existing models that were recently validated on the same databases. Therefore,

ensemble classification system can be employed to increase the reliability and consis-

tency of financial data classification task.

KEYWORDS

bankruptcy prediction, credit scoring, ensemble classifiers, ensemble learning, financial data

classification

1|INTRODUCTION

An ensemble learning and classification system is composed of multi-

ple classifiers or sub-systems. The goal is to improve accuracy (reduce

prediction error) by combining responses produced by these multiple

classifiers into a single response or output. In this framework, since

the final output is computed by combining outputs from different

classifiers, the final ensemble system decision generally is wrong only

when most of the sub-systems make the same error.

In general, an ensemble learning and classification system is

homogeneous when it is composed of classifiers trained with the

same learning algorithm, whereas it is heterogeneous when it is

composed of classifiers trained with different learning algorithms.

For instance, bagging and boosting algorithms are commonly used

to build homogeneous ensembles with various samples or parame-

ters to build base classifiers. In contrary, the stacking algorithm

generally is employed to generate heterogeneous ensemble sys-

tems. Because of their good performance, ensemble learning and

classification systems have been successfully employed in various

applications; including customer churn prediction (Xiao, Jiang, He, &

Teng, 2016), person authentication (Pisani, Lorena & de Carvalho,

2018), visual categorization (Zhu, Shao, & Fang, 2016), microblog

summarization (Dutta et al., 2018), sentiment classification (Xia,

Zong, Hu & Cambria, 2013), energy price time series prediction

(Lahmiri, 2017a), and stock market forecasting (Lahmiri, 2018;

Lahmiri & Boukadoum, 2015).

Received: 10 May 2019 Revised: 18 October 2019 Accepted: 18 October 2019

DOI: 10.1002/isaf.1460

Intell Sys Acc Fin Mgmt. 2020;27:3–9. wileyonlinelibrary.com/journal/isaf © 2020 John Wiley & Sons, Ltd. 3