Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fair and Actionable Causal Prescription Ruleset

Benton Li, Nativ Levy, Brit Youngmann, Sainyam Galhotra, Sudeepa Roy

TL;DR

This paper addresses generating fair, causal prescriptions (rulesets) to improve outcomes without widening inequalities across protected groups. It introduces FairCap, a three-step algorithm that combines Apriori-based grouping pattern mining, lattice-based search for fair interventions, and greedy rule selection to efficiently explore a vast search space under fairness and coverage constraints. The framework formalizes prescription rules, coverage, utility via $\mathrm{ExpUtility}$, and two fairness notions (statistical parity and bounded group loss), proving NP-hardness and showing matroid-structured variants amenable to greedy methods. Empirical case studies on German Credit and Stack Overflow demonstrate that incorporating fairness reduces disparities at some cost to total utility, while the approach remains scalable to large datasets and adaptable across problem variants.

Abstract

Prescriptions, or actionable recommendations, are commonly generated across various fields to influence key outcomes such as improving public health, enhancing economic policies, or increasing business efficiency. While traditional association-based methods may identify correlations, they often fail to reveal the underlying causal factors needed for informed decision-making. On the other hand, in decision-making for tasks with significant societal or economic impact, it is crucial to provide recommendations that are justifiable and equitable in terms of the outcome for both the protected and non-protected groups. Motivated by these two goals, this paper introduces a fairness-aware framework leveraging causal reasoning for generating a set of actionable prescription rules (ruleset) toward betterment of an outcome while preventing exacerbating inequalities for protected groups. By considering group and individual fairness metrics from the literature, we ensure that both protected and non-protected groups benefit from these recommendations, providing a balanced and equitable approach to decision-making. We employ efficient optimizations to explore the vast and complex search space considering both fairness and coverage of the ruleset. Empirical evaluation and case study on real-world datasets demonstrates the utility of our framework for different use cases.

Fair and Actionable Causal Prescription Ruleset

TL;DR

This paper addresses generating fair, causal prescriptions (rulesets) to improve outcomes without widening inequalities across protected groups. It introduces FairCap, a three-step algorithm that combines Apriori-based grouping pattern mining, lattice-based search for fair interventions, and greedy rule selection to efficiently explore a vast search space under fairness and coverage constraints. The framework formalizes prescription rules, coverage, utility via , and two fairness notions (statistical parity and bounded group loss), proving NP-hardness and showing matroid-structured variants amenable to greedy methods. Empirical case studies on German Credit and Stack Overflow demonstrate that incorporating fairness reduces disparities at some cost to total utility, while the approach remains scalable to large datasets and adaptable across problem variants.

Abstract

Prescriptions, or actionable recommendations, are commonly generated across various fields to influence key outcomes such as improving public health, enhancing economic policies, or increasing business efficiency. While traditional association-based methods may identify correlations, they often fail to reveal the underlying causal factors needed for informed decision-making. On the other hand, in decision-making for tasks with significant societal or economic impact, it is crucial to provide recommendations that are justifiable and equitable in terms of the outcome for both the protected and non-protected groups. Motivated by these two goals, this paper introduces a fairness-aware framework leveraging causal reasoning for generating a set of actionable prescription rules (ruleset) toward betterment of an outcome while preventing exacerbating inequalities for protected groups. By considering group and individual fairness metrics from the literature, we ensure that both protected and non-protected groups benefit from these recommendations, providing a balanced and equitable approach to decision-making. We employ efficient optimizations to explore the vast and complex search space considering both fairness and coverage of the ruleset. Empirical evaluation and case study on real-world datasets demonstrates the utility of our framework for different use cases.

Paper Structure

This paper contains 27 sections, 4 theorems, 9 equations, 5 figures, 6 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. Background on Causal Inference
  4. Problem Formulation
  5. Pattern and Protected Group
  6. Prescription Rules
  7. Utility of a Prescription Ruleset
  8. Size of a Prescription Ruleset
  9. Coverage Constraints
  10. Fairness Constraints
  11. Statistical parity
  12. Bounded group loss (BGL)
  13. The Prescription Ruleset Selection Problem
  14. The FairCap Algorithm
  15. Step 1: Mining Grouping Patterns
  16. ...and 12 more sections

Key Result

lemma 1

Given a rule $r {=} (\mathcal{P}_{{\tt grp}}, \mathcal{P}_{{\tt int}})$, there exists a rule $r' {=} (\mathcal{P}_{g'}, \mathcal{P}_{t'})$ s.t $\mathcal{P}_{g'} {\subset} \textsc{Coverage}(\mathcal{P}_{{\tt grp}})$ and $utility(r') {\geq} utility(r)$.

Figures (5)

  • Figure 1: Partial causal DAG for the Stack Overflow dataset.
  • Figure 2: A decision tree for selecting the appropriate problem variant.
  • Figure 3: Runtime by-step of the FairCap algorithm (SO)
  • Figure 4: Runtime as a function of the dataset size (SO)
  • Figure 5: Runtime as a function of number of mutable and immutable attributes for SO with statistical parity

Theorems & Definitions (16)

  • Example 1.1
  • Example 1.2
  • Example 1.3
  • Example 3.1
  • Definition 4.1: Pattern
  • Example 4.1
  • Definition 4.2: Coverage of a pattern
  • Definition 4.3: Prescription Rule and Ruleset, Grouping and intervention patterns, and Coverage
  • Example 4.2
  • Definition 4.4: Utility of a prescription rule - overall, protected, non-protected
  • ...and 6 more