Three-MLP Ensemble Re-RX algorithm and recent classifiers for credit-risk evaluation

Haysahi, Yoichi (Departement of Computer Science, Meiji University, Tama-ku, Kawasaki, Japan) ; Tanaka, Yuki (Departement of Computer Science, Meiji University, Tama-ku, Kawasaki, Japan) ; Yukita, Shonosuke (Departement of Computer Science, Meiji University, Tama-ku, Kawasaki, Japan) ; Nakano, Satoshi (Departement of Computer Science, Meiji University, Tama-ku, Kawasaki, Japan) ; Bologna, Guido (School of Engineering, Architecture and Landscape (hepia), HES-SO // University of Applied Sciences Western Switzerland)

Credit-risk evaluation is a challenging and important task in the domain of financial analysis for which many classification methods have been suggested. In this paper, we present the results for eight real-life credit-risk two-class mixed datasets (i.e., discrete and continuous attributes) analyzed by the Three-MLP Ensemble Re-RX algorithm (shortened to “Three-MLP Ensemble”). Clarifying the neural network decisions by explanatory rules that capture the learned knowledge embedded in the networks can help a credit-risk manager explain why a particular applicant is classified as either bad or good. To compare the Three-MLP Ensemble performance, we executed comprehensive rule extraction experiments on eight two-class mixed datasets commonly used for benchmarking studies in credit-risk evaluation. The extremely high accuracy of the Three-MLP Ensemble outperformed the accuracies by the Re-RX algorithm and a variant. In this study, we also compared the accuracy of the Three-MLP Ensemble with that of classifiers recently proposed. It is concluded that neural network rule extraction by the Three-MLP Ensemble is a powerful management tool that allows us to build advanced, comprehensible, and accurate decision-support systems for credit-risk evaluation.

Conference Type:
full paper
Ingénierie et Architecture
HEPIA - Genève
inIT - Institut d'Ingénierie Informatique et des Télécommunications
Killarney, Ireland, 12-17 July 2015
Killarney, Ireland
12-17 July 2015
8 p.
Published in:
Proceedings of 2015 Inernational Joint Conference on neural Networks (IJCNN), 12-17 July 2015, Killarney, Ireland
Appears in Collection:

Note: The status of this file is: restricted

 Record created 2020-02-21, last modified 2020-02-28

Download fulltext

Rate this document:

Rate this document:
(Not yet reviewed)