Ensemble of Classifiers with Dynamic Update for Credit Risk Analysis

• Grant number: 13/11615-6
• Support Opportunities: Scholarships in Brazil - Doctorate
• Start date: November 01, 2013
• End date: January 09, 2018
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Principal Investigator: André Carlos Ponce de Leon Ferreira de Carvalho
• Grantee: Everlandio Rebouças Queiroz Fernandes
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Associated scholarship(s): 16/20465-6 - Ensemble of Classiers for Unbalanced Data sets, BE.EP.DR 15/01370-1 - Ensemble of classifiers for unbalanced datasets, BE.EP.DR

Abstract

In credit analysis applications, the databases are often heavily unbalanced, presenting a much larger number of examples from the negative class (complaint clients) than from the positive class (default clients). Unbalanced Databases compromise the performance of most classical classification algorithms because they assume a balanced distribution of examples between the classes and consider the cost of misclassification to be the same for all classes. Some strategies have been proposed to overcome this problem, but they may not improve the predictive performance for some databases. Another important feature of these databases is that they are constantly updated, i.e. recently inserted clients in the database should be labeled as compliant or default, leading, in general, to the need of updating the model. This feature, observed in data streams, can be used to identify the emergence of new classes and new trends in the data. In many classification problems, the combination of more than one classifier in structures, known as ensemble of classifiers, has shown predictive accuracy stable and often superior to the use of a simple classifier. However, the problems investigated are usually restricted to static data sets. It is proposed in this project the investigation and the proposal of a new approach for ensemble of classifiers based on evolutionary algorithms. These algorithms will be used to select examples from the training database so that the classification model induced can return classifiers with higher predictive accuracy. The project will alsoinvestigate new approaches to keep the model updated with the arrival of new data, already labeled, often inserted in credit analysis databases.

Scientific publications

(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)

FERNANDES, EVERLANDIO; ROCHA, RAFAEL L.; FERREIRA, BRUNO; CARVALHO, EDUARDO; SIRAVENHA, ANA CAROLINA; GOMES, ANA CLAUDIA S.; CARVALHO, SCHUBERT; DE SOUZA, CLEIDSON R. B.; IEEE. An Ensemble of Convolutional Neural Networks for Unbalanced Datasets: A case Study with Wagon Component Inspection. 2018 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS (IJCNN), v. N/A, p. 6-pg., 2018-01-01. (13/11615-6)

Academic Publications

(References retrieved automatically from State of São Paulo Research Institutions)

FERNANDES, Everlandio Rebouças Queiroz. Evolutionary ensembles for imbalanced learning. 2018. Doctoral Thesis - Universidade de São Paulo (USP). Instituto de Ciências Matemáticas e de Computação (ICMC/SB) São Carlos.