Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The more the merrier: logical and multistage processors in credit scoring

Arturo Pérez-Peralta, Sandra Benítez-Peña, Rosa E. Lillo

TL;DR

This paper addresses fairness in credit scoring by introducing two methods that extend fairness beyond single-attribute settings: logical processors (LP) that map multiple sensitive attributes into one binary variable, and multistage processors (MP) that stack fairness methods across pre-, in-, and post-processing pipeline stages. The authors formalize fairness criteria (Independence, Separation, Sufficiency) and argue that, for credit scoring, separation is often the most appropriate metric, then implement and evaluate LPs and MPs across simulations and a German loan dataset. Empirical results show LPs can effectively handle multiple sensitive attributes with fairness and accuracy preserved, while MPs can yield synergistic improvements—sometimes surpassing individual methods—in both accuracy and separation, albeit with dependencies on the LP choice and post-processing stages. The findings suggest practical pathways to enforce multi-attribute fairness in finance and point to future work on expanding the processor repertoire, enhancing interpretability, and tuning hyper-parameters for robust deployment.

Abstract

Machine Learning algorithms are ubiquitous in key decision-making contexts such as organizational justice or healthcare, which has spawned a great demand for fairness in these procedures. In this paper we focus on the application of fair ML in finance, more concretely on the use of fairness techniques on credit scoring. This paper makes two contributions. On the one hand, it addresses the existent gap concerning the application of established methods in the literature to the case of multiple sensitive variables through the use of a new technique called logical processors (LP). On the other hand, it also explores the novel method of multistage processors (MP) to investigate whether the combination of fairness methods can work synergistically to produce solutions with improved fairness or accuracy. Furthermore, we examine the intersection of these two lines of research by exploring the integration of fairness methods in the multivariate case. The results are very promising and suggest that logical processors are an appropriate way of handling multiple sensitive variables. Furthermore, multistage processors are capable of improving the performance of existing methods.

The more the merrier: logical and multistage processors in credit scoring

TL;DR

This paper addresses fairness in credit scoring by introducing two methods that extend fairness beyond single-attribute settings: logical processors (LP) that map multiple sensitive attributes into one binary variable, and multistage processors (MP) that stack fairness methods across pre-, in-, and post-processing pipeline stages. The authors formalize fairness criteria (Independence, Separation, Sufficiency) and argue that, for credit scoring, separation is often the most appropriate metric, then implement and evaluate LPs and MPs across simulations and a German loan dataset. Empirical results show LPs can effectively handle multiple sensitive attributes with fairness and accuracy preserved, while MPs can yield synergistic improvements—sometimes surpassing individual methods—in both accuracy and separation, albeit with dependencies on the LP choice and post-processing stages. The findings suggest practical pathways to enforce multi-attribute fairness in finance and point to future work on expanding the processor repertoire, enhancing interpretability, and tuning hyper-parameters for robust deployment.

Abstract

Machine Learning algorithms are ubiquitous in key decision-making contexts such as organizational justice or healthcare, which has spawned a great demand for fairness in these procedures. In this paper we focus on the application of fair ML in finance, more concretely on the use of fairness techniques on credit scoring. This paper makes two contributions. On the one hand, it addresses the existent gap concerning the application of established methods in the literature to the case of multiple sensitive variables through the use of a new technique called logical processors (LP). On the other hand, it also explores the novel method of multistage processors (MP) to investigate whether the combination of fairness methods can work synergistically to produce solutions with improved fairness or accuracy. Furthermore, we examine the intersection of these two lines of research by exploring the integration of fairness methods in the multivariate case. The results are very promising and suggest that logical processors are an appropriate way of handling multiple sensitive variables. Furthermore, multistage processors are capable of improving the performance of existing methods.

Paper Structure

This paper contains 23 sections, 18 equations, 14 figures, 4 tables.

Table of Contents

  1. Introduction
  2. The Foundations of Fairness
  3. The Classification problem under the fair pointwise
  4. Fairness criteria
  5. Independence
  6. Separation
  7. Sufficiency
  8. What metric to use?
  9. Fairness processors
  10. Pre-Processing
  11. In-Processing
  12. Post-Processing
  13. Extending fairness
  14. Multiple sensitive variables and logical processors
  15. Multistage processors
  16. ...and 8 more sections

Figures (14)

  • Figure 1: Accuracy of multistage processors in the simulation study. The main diagonal shows the accuracy of each method while the off-diagonal elements represent the performance of the combination of two processors.
  • Figure 2: Accuracy graph of multistage processors for the simulation study. Nodes show the accuracy of individual methods while edges represent the performance of the combination of two methods.
  • Figure 3: Separation heatmap for the simulation study.
  • Figure 4: Separation graph for the simulation study
  • Figure 5: Boxplots of the results of the simulation study for selected methods. Ten individual instances selected are random are highlighted with different color for easier tracing.
  • ...and 9 more figures