Paper: arXiv 2512.24747

Authors: Tim J. Boonen, Xinyue Fan, Zixiao Quan

Abstract

Machine learning improves predictive accuracy in insurance pricing but exacerbates trade-offs between competing fairness criteria across different discrimination measures, challenging regulators and insurers to reconcile profitability with equitable outcomes. While existing fairness-aware models offer partial solutions under GLM and XGBoost estimation methods, they remain constrained by single-objective optimization, failing to holistically navigate a conflicting landscape of accuracy, group fairness, individual fairness, and counterfactual fairness. To address this, we propose a novel multi-objective optimization framework that jointly optimizes all four criteria via the Non-dominated Sorting Genetic Algorithm II (NSGA-II), generating a diverse Pareto front of trade-off solutions. We use a specific selection mechanism to extract a premium on this front. Our results show that XGBoost outperforms GLM in accuracy but amplifies fairness disparities; the Orthogonal model excels in group fairness, while Synthetic Control leads in individual and counterfactual fairness. Our method consistently achieves a balanced compromise, outperforming single-model approaches.

Complexity vs Empirical Score

  • Math Complexity: 6.5/10
  • Empirical Rigor: 7.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper presents a novel multi-objective framework (using NSGA-II and TOPSIS) to navigate complex trade-offs between predictive accuracy and multiple fairness definitions (group, individual, counterfactual) in insurance pricing. It balances solid mathematical definitions of fairness metrics with practical empirical evaluation across standard machine learning and actuarial estimation techniques.

Research Flowchart

  flowchart TD
    A[Research Goal: Balance Accuracy, Group, Individual, & Counterfactual Fairness in Insurance Pricing] --> B[Data Inputs: Synthetic Control & Orthogonal Model Baselines under GLM vs XGBoost]
    B --> C[Methodology: Multi-Objective Optimization Framework using NSGA-II]
    C --> D[Computational Process: Jointly Optimize 4 Criteria to Generate Diverse Pareto Front]
    D --> E[Selection Mechanism: Extract Optimal Balanced Premium from Pareto Front]
    E --> F[Findings: XGBoost Increases Accuracy but Disparities; NSGA-II Outperforms Single-Models]

Cited References (103)

  1. The discriminating (pricing) actuary (2023)
  2. The frontiers of fairness in machine learning (2018) — link
  3. A review on fairness in machine learning (2022)
  4. Fair regression: Quantitative definitions and reduction-based algorithms (2019)
  5. Fairness in Machine Learning: A Survey.(2020) (2010) — link
  6. Equality of opportunity in supervised learning (2016)
  7. Algorithmic decision making and the cost of fairness (2017)
  8. Antidiscrimination insurance pricing: Regulations, fairness criteria, and models (2024)
and 95 more
  1. From actuarial to behavioural valuation. the impact of telematics onmotor insurance (2022)
  2. Community rating in the absence of risk equalisation: lessons from the Irish private health insurance market (2013)
  3. A unisex stochastic mortality model to comply with EU Gender Directive (2017)
  4. Fairness through awareness (2012)
  5. Fairness through computationally-bounded awareness (2018)
  6. Accurate fairness: Improving individual fairness without trading accuracy (2023)
  7. Counterfactual fairness (2017)
  8. Fairness in criminal justice risk assessments: The state of the art (2021)
  9. Counterfactual fairness: Unidentification, bound and algorithm (2019)
  10. Causality (2009)
  11. Discrimination-free insurance pricing (2022)
  12. Path-specific counterfactual fairness (2019)
  13. Counterfactual fairness by combining factual and counterfactual predictions (2024)
  14. Avoiding discrimination through causal reasoning (2017)
  15. A fair price to pay: Exploiting causal graphs for fairness in insurance (2025)
  16. Implementing anti-discrimination policies in statistical profiling models (2011)
  17. Exposing the probabilistic causal structure of discrimination (2017)
  18. Individual fairness under uncertainty (2023) — link
  19. Inform: Individual fairness on graph mining (2020)
  20. Post-processing for individual fairness (2021)
  21. A general coefficient of similarity and some of its properties (1971)
  22. Aligning relational learning with lipschitz fairness (2024)
  23. Estimation and inference of heterogeneous treatment effects using random forests (2018)
  24. Estimating treatment effects with causal forests: An application (2019)
  25. Combining predictions of auto insurance claims (2022)
  26. Welfare Implications of Fairness Regulations in Insurance Cost Modeling: A Multi-Method Study (2025)
  27. Certifying and removing disparate impact (2015)
  28. Insurance, biases, discrimination and fairness (2024)
  29. Fair regression with wasserstein barycenters (2020)
  30. Obtaining fairness using optimal transport theory (2019)
  31. Synthetic control methods for comparative case studies: Estimating the effect of C alifornia’s tobacco control program (2010)
  32. Robust synthetic control (2018)
  33. The augmented synthetic control method (2021)
  34. CASdatasets: Insurance datasets (2025) — link
  35. XGBoost : A scalable tree boosting system (2016)
  36. The accuracy of XGBoost for insurance claim prediction (2018)
  37. Techniques for discrimination-free predictive models (2013)
  38. Fairness-aware training of decision trees by abstract interpretation (2021)
  39. Individually Fair Gradient Boosting (2021)
  40. Fair neighbor embedding (2023)
  41. Towards a unified framework for fair and stable graph representation learning (2021)
  42. Don’t let your analysis go to seed: on the impact of random seed on machine learning-based causal inference (2024)
  43. Counterfactual fairness: removing direct effects through regularization (2020) — link
  44. Generalized linear models for insurance rating (2016)
  45. A review of multi-objective optimization: Methods and its applications (2018)
  46. Multi-objective optimization using genetic algorithms: A tutorial (2006)
  47. Multiple-gradient descent algorithm (MGDA) for multiobjective optimization (2012)
  48. Fairness constraints: Mechanisms for fair classification (2017)
  49. A fast and elitist multiobjective genetic algorithm: NSGA-II (2002)
  50. Nonlinear multiobjective optimization (1999)
  51. Recycling privileged learning and distribution matching for fairness (2017)
  52. A human-in-the-loop fairness-aware model selection framework for complex fairness objective landscapes (2024)
  53. Combining forecasts using meta-learning: a comparative study for complex seasonality (2023)
  54. Model-agnostic meta-learning for fast adaptation of deep networks (2017)
  55. A state-of the-art survey of TOPSIS applications (2012)
  56. Enhanced Structural Design of Prestressed Arched Trusses through Multi-Objective Optimization and Multi-Criteria Decision-Making (2024)
  57. Boosting insights in insurance tariff plans with tree-based machine learning methods (2021)
  58. Fundamentals of actuarial mathematics (2014)
  59. Treatment of Profit and Contingency Provisions and the Cost of Capital in Property/Casualty Insurance Ratemaking (1997) — link
  60. Property and Casualty Model Rating Law (File and Use Version) (2010) — link
  61. Council Directive 2004/113/EC – Implementing the Principle of Equal Treatment between Men and Women in the Access to and Supply of Goods and Services (2004)
  62. Proxy discrimination in the age of artificial intelligence and big data (2019)
  63. Machine Learning with Multitype Protected Attributes: Intersectional Fairness through Regularisation — → 2509.08163
  64. Limits and potential of the concept of indirect discrimination (2008)
  65. Discrimination and insurance (2017)
  66. Data-Driven Discrimination, Perceived Fairness, and Consumer Trust–The Perspective of Consumer Attribution (2021)
  67. Born free and equal?: A philosophical inquiry into the nature of discrimination (2013)
  68. Big data’s disparate impact (2016)
  69. Big data and actuarial science (2020)
  70. Defining discrimination in insurance (2022)
  71. The impact of telematics on the insurability of risks (2020)
  72. Artificial intelligence and personalization of insurance: Failure or delayed ignition? (2025)
  73. AI revolution in insurance: bridging research and reality (2025)
  74. The Genetic Information Nondiscrimination Act (GINA ): public policy and medical practice in the age of personalized medicine (2012)
  75. Methods for quantifying discriminatory effects on protected classes in insurance (2021)
  76. Adverse impact and test validation: A practitioner’s guide to valid and defensible employment testing (2017)
  77. Private health insurance in the European Union (2009)
  78. Does Limiting Allowable Rating Variation in the Small Group Health Insurance Market Affect Employer Self-Insurance? (2018)
  79. Price optimisation for insurance optimising price; destroying value (2016)
  80. FairBranch: Mitigating Bias Transfer in Fair Multi-task Learning (2024)
  81. Counterfactual fairness for facial expression recognition (2022)
  82. When is discrimination wrong? (2008)
  83. Beyond distributive fairness in algorithmic decision making: Feature selection for procedurally fair learning (2018)
  84. When machine learning meets privacy: A survey and outlook (2021)
  85. Deferred annuities with gender-neutral pricing: Benefitting most women without adversely affecting too many men (2024)
  86. Fairness-Aware Neural R'eyni Minimization for Continuous Features (2019) — link
  87. Fairness definitions explained (2018) — link
  88. Do the ends justify the means? Variation in the distributive and procedural fairness of machine learning algorithms (2022)
  89. Equality of opportunity (2015)
  90. Methods for multiple attribute decision making (1981)
  91. Comparison of weights in TOPSIS models (2004)
  92. Multi-objective optimization for customized solar business models considering technical-economic-environmental performance: A NSGA-II integrated TOPSIS method (2025)
  93. New decision-making methods for ranking of non-dominated points for multi-objective optimization problems (2024)
  94. A comprehensive review on NSGA-II for multi-objective combinatorial optimization problems (2021)
  95. An improved NSGA-II algorithm based on crowding distance elimination strategy (2019)

1 of 103 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.