Explainable AI for Credit Assessment in Banks

Round 1

Reviewer 1 Report

The paper is well-written, structurally organized, and very clear and detailed in explanations. However, a linguistic revision is required due to minor spelling and construction errors.

The paper exploits a comprehensive granular database on consumer loans and presents a paper based on an explainability artificial intelligence (XAI) algorithm based on Gradient Boosting Model (GBM). In this sense, XAI algorithms are able to analyze how each variable affects credit default predictions. Furthermore, the authors provide a banking application of XAI algorithms based on real-life loan data, and they quantify the model accuracy improvements with respect to logistic regression models.

The paper contributes to the literature by providing an empirical application of XAI algorithms to an actual customer credit card database. I suggest integrating the literature with (at least) the following papers:

a) Bussmann N., Giudici P., Marinelli D., Papenbrock J., (2020), “Explainable Machine Learning in Credit Risk Management”, Computational Economics, 57, pp. 203–216. https://doi.org/10.1007/s10614-020-10042-0;

b) Moscato V., Picariello A., Sperlí G., (2021), “A benchmark of machine learning approaches for credit score prediction“, Expert Systems With Applications, 165, 113986. https://doi.org/10.1016/j.eswa.2020.113986

The application of Light Gradient Boosting Machine (LightGBM) models allows the reduction of the workload by the algorithm without precluding the accuracy of the results.

The authors described the methodological steps clearly and in detail regarding the Gradient Boosting Decision Trees. Regarding the logistic regression model, it would be appropriate to explain the regression model in the methodological paragraph.

The division of methodological results from economic interpretation is a strong point of the paper. The application aimed at identifying the LGD exploiting XAI algorithms is a theme of interest for credit institutions and supervisory entities.

The use of XAI algorithms for credit risk analysis is clearly explained in the conclusions. The authors highlight the weaknesses of evaluation models based on logistical regressions and propose an alternative model based on artificial intelligence but with more easily interpretable results.

Summing up:

1. The author needs to provide a statistic summary for the dataset.

2. There is no information about further developments in the research field.

3. The authors need to explain the limitations of this study in the conclusions.

4. A correlation matrix and a list of variables should be presented.

5. Authors should enrich the literature review.

6. Explicit the logistic model used.

7. Minor English revision.

Author Response

See attachment

Author Response File: Author Response.docx

Reviewer 2 Report

Must include some results in the abstract section.

There must be problem statement and objectives modules under the heading of the Introduction for a better understanding of the proposed research.

Please include the Discussion by comparing the proposed and existing approaches along with their reference.

Please elaborate conclusion and must be relevant to the proposed research.

Author Response

See attachment

Author Response File: Author Response.docx