EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

4396. Deep Learning for Credit Scoring: Do or Don’t – A retrospective look

Invited abstract in session MD-8: EJOR: policy, facts and highlights, stream OR Journals.

Monday, 14:30-16:00
Room: 1020 (building: 202)

Authors (first author is the speaker)

1. Bart Baesens
2. María Óskarsdóttir
Department of Computer Science, Reykjavik University
3. Wilfried Lemahieu
Faculty of Economics and Business, KU Leuven
4. Björn Rafn Gunnarsson
KU Leuven
5. Seppe vanden Broucke
Department of Decision Sciences and Information Management, KU Leuven

Abstract

New technologies such as deep learning should always be approached with scientific rigor. Using a unique and extensive collection of tabular data sets, we wanted to explore the potential of deep learning for credit scoring. We believe one of the key novelties of this paper is the robust empirical comparison framework we set up, using multiple datasets, evaluation metrics, confidence intervals, frequentist and Bayesian testing of significance. What we often encounter when reviewing papers, is that researchers either don’t give baseline methods a fair chance or cherry pick them when comparing against their own newly proposed method(s). Furthermore, something rather basic but not unimportant relates to the title and message of the paper. We opted for a simple, easily Google-able title properly reflecting the research question which not only draws attention but also allowed readers to very easily grasp the key takeaways of our work. In this presentation, we elaborate more on this as well as how this paper helped shape our future research.

Keywords

Status: accepted

Back to the list of papers