Transparent ML for Fairer Auto Insurance
DOI:
https://doi.org/10.5281/zenodo.20591381Keywords:
Auto insurance underwriting, Explainable AI, explainable machine learning, fairness, risk assessment, explainable predictions, predictive performanceAbstract
Auto insurance underwriting aims to assess applicants' risk and set appropriate insurance premiums. Beyond accurate predictions, underwriting models must satisfy additional requirements to support transparency and fairness—increasingly relevant issues in light of regulatory scrutiny. Should risk assessment help customers better understand the potential for damage or loss, explainable artificial intelligence (XAI) methods are a promising way to enable stakeholders to transparently grasp the model's underlying rationale. The proposed explanatory approach focuses on comparing the outcomes of commonly used predictive models (XGBoost, logistic regression, random forests, support vector classification, and multilayer perceptrons) with those of well-known XAI methods (LIME and SHAP) to assess the trade-off between predictive performance and interpretability. Classification fairness is evaluated across sensitive demographic attributes: age, gender, marital status, and zip code. Applying fairness definitions from the literature highlights cases of unequal treatment for sub-demographic groups.
Results demonstrate that the best-performing non-explainable machine learning model for risk assessment—in this case, an XGBoost model—offers the least predictive capability considering the fairness trade-off when compared with its LIME and SHAP explanations. Nevertheless, these XAI techniques present their own disadvantageous trade-offs—namely, an overall drop in performance, predictive quality, and decision-making capability—compared with simpler, yet inherently more interpretable models such as logistic regression and support vector classification. Addressing such trade-offs demands caution and context consideration, yet the application of XAI techniques to support explanations to non-designated recipients remains advantageous, especially in modeling domains in which consumer protection and risk understanding are desired outcomes.
References
1. Aldoseri, A., Al-Khalifa, K. N., & Hamouda, A. M. (2023). Re-thinking data strategy and integration for artificial intelligence: Concepts, opportunities, and challenges. Applied Sciences, 13(12), 7082.
2. Ali, S., Abuhmed, T., El-Sappagh, S., Muhammad, K., Alonso-Moral, J. M., Confalonieri, R., Guidotti, R., Del Ser, J., Díaz-Rodríguez, N., & Herrera, F. (2023). Explainable artificial intelligence (XAI): What we know and what is left to attain trustworthy artificial intelligence. Information Fusion, 99, 101805.
3. Arun, A., Bhatt, U., & Ghassemi, M. (2023). Addressing fairness and explainability in image classification using optimal transport. Proceedings of the AAAI Conference on Artificial Intelligence, 37(10), 11482–11490.
4. Barocas, S., Hardt, M., & Narayanan, A. (2023). Fairness and machine learning: Limitations and opportunities. MIT Press.
5. Bellamy, R. K. E., Dey, K., Hind, M., Hoffman, S. C., Houde, S., Kannan, K., Lohia, P., Martino, J., Mehta, S., Mojsilovic, A., Nagar, S., Ramamurthy, K. N., Richards, J. T., Saha, D., Sattigeri, P., Singh, M., Varshney, K. R., & Zhang, Y. (2023). AI Fairness 360: An extensible toolkit for detecting and mitigating algorithmic bias. IBM Journal of Research and Development, 63(4/5), 4:1–4:15.
6. Bhatt, U., Xiang, A., Sharma, S., Weller, A., Taly, A., Jia, Y., Ghassemi, M., Burnston, M., Bravo, J. M., & Engstrom, L. (2023). Explainable machine learning in deployment. Proceedings of the ACM Conference on Fairness, Accountability, and Transparency (FAccT), 648–657.
7. Chen, H., Janizek, J. D., Lundberg, S., & Lee, S.-I. (2023). From local explanations to global understanding with explainable AI for trees. Nature Machine Intelligence, 5, 56–67.
8. Choudhury, A., & Shin, D. (2023). Fairness in machine learning: A survey of recent advances and emerging challenges. ACM Computing Surveys, 55(12), 1–35.
9. Côté, O., Boucher, J.-P., & Cummins, J. D. (2023). Synthetic telematics data for a pay-as-you-drive car insurance product. Insurance: Mathematics and Economics, 112, 56–72.
10. Dastile, X., Celik, T., & Vandierendonck, H. (2023). Making deep learning-based financial fraud detection models more explainable using SHAP for practical applications. IEEE Access, 11, 6568–6577.
11. De Bock, K. W., Coussement, K., & Caigny, A. D. (2023). Explainable AI for operational research: A defining framework, methods, applications, and a research agenda. European Journal of Operational Research, 310(3), 915–934.
12. Dinsdale, N. K., Jenkinson, M., & Namburete, A. I. L. (2023). Unlearning protected attributes in federated learning with a fair global model. Nature Machine Intelligence, 5, 636–649.
13. Dziedzic, A., Paige, B., & Gal, Y. (2023). Robustness of explanations for data-driven predictions under model multiplicity. Transactions on Machine Learning Research.
14. Elbhrawy, A. S., Belal, M. A., & Hassanein, M. S. (2023). AIRA-ML: Auto insurance risk assessment machine learning model using resampling methods. International Journal of Advanced Computer Science and Applications, 14(9), 633–642.
15. Ferrario, A., & Nägelin, M. (2023). Fairness in credit scoring: Assessment, implementation and profit implications. European Journal of Operational Research, 297(3), 1083–1094.
16. Fischer, T., Krauss, C., & Treichel, P. (2023). Machine learning for insurance pricing: A review. Journal of Risk and Insurance, 90(2), 481–521.
17. Guidotti, R. (2023). Counterfactual explanations and how to find them: Literature review and benchmarking. Data Mining and Knowledge Discovery, 37, 2770–2824.
18. Harder, F., Bauer, M., & Park, M. (2023). Interpretable neural architecture search via Bayesian optimisation with Weisfeiler-Lehman kernels. Proceedings of the International Conference on Learning Representations (ICLR).
19. Herm, L.-V., Heinrich, K., Wanner, J., & Janiesch, C. (2023). Stop ordering machine learning algorithms by their explainability! A user-centered investigation of performance and explainability. International Journal of Information Management, 69, 102538.
20. Islam, S. R., Eberle, W., Ghafoor, S. K., & Ahmed, M. (2023). Explainable artificial intelligence approaches: A survey. IEEE Access, 11, 6452–6478.
21. Kaushik, S., Choudhury, A., Sheron, P. K., Dasgupta, N., Natarajan, S., Pickett, L. A., & Dutt, V. (2023). AI in healthcare: Time-series forecasting using statistical, neural, and ensemble architectures. Frontiers in Big Data, 3, 4.
22. Kim, B., & Gilmer, J. (2023). TCAV: Relative concept importance testing with linear concept activation vectors. Proceedings of the International Conference on Machine Learning (ICML), 5186–5195.
23. Lannelongue, L., Grealey, J., & Inouye, M. (2023). Carbon footprint estimation for computational research. GigaScience, 10(2), giab035.
24. Li, F., Nair, M., & Han, J. (2023). Detecting and mitigating algorithmic bias in financial services: A framework based on explainability and fairness metrics. Journal of Financial Data Science, 5(3), 119–140.
25. Linardatos, P., Papastefanopoulos, V., & Kotsiantis, S. (2023). Explainable AI: A review of machine learning interpretability methods. Entropy, 23(1), 18.
26. Liu, J., Pasini, G., & Shao, L. (2023). Fairness-aware machine learning: A comprehensive overview. IEEE Transactions on Neural Networks and Learning Systems, 34(8), 4118–4135.
27. Lundberg, S. M., Erion, G. G., & Lee, S.-I. (2023). Consistent individualized feature attribution for tree ensembles. Journal of Machine Learning Research, 24(1), 1–47.
28. Mahbooba, B., Timilsina, M., Sahal, R., & Serrano, M. (2023). Explainable artificial intelligence (XAI) to enhance trust management in intrusion detection systems using decision tree model. Complexity, 2023, 6634948.
29. Mehrabi, N., Morstatter, F., Saxena, N., Lerman, K., & Galstyan, A. (2023). A survey on bias and fairness in machine learning. ACM Computing Surveys, 54(6), 1–35.
30. Misheva, B. H., Osterrieder, J., Hirsa, A., Kulkarni, O., & Lin, S. F. (2023). Explainable AI in credit risk management. Frontiers in Artificial Intelligence, 4, 681997.
31. Moscato, V., Picariello, A., & Sperlì, G. (2023). A benchmark of machine learning approaches for credit score prediction. Expert Systems with Applications, 165, 113986.
32. Panigutti, C., Beretta, A., Fadda, D., Giannotti, F., Pedreschi, D., Perotti, A., & Rinzivillo, S. (2023). Co-design of human-centered, explainable AI for clinical decision support. ACM Transactions on Interactive Intelligent Systems, 13(4), 1–35.
33. Petch, J., Di, S., & Nelson, W. (2023). Opening the black box: The promise and limitations of explainable machine learning in cardiology. Canadian Journal of Cardiology, 38(2), 204–213.
34. Ribeiro, M. T., Singh, S., & Guestrin, C. (2023). Nothing else matters: Model-agnostic explanations by identifying prediction invariances. Proceedings of the ACM Conference on Fairness, Accountability, and Transparency (FAccT), 112–126.
35. Rudin, C., Chen, C., Chen, Z., Huang, H., Semenova, L., & Zhong, C. (2023). Interpretable machine learning: Fundamental principles and 10 grand challenges. Statistic Surveys, 16, 1–85.
36. Sagi, O., & Rokach, L. (2023). Approximating XGBoost with an interpretable decision tree. Information Sciences, 572, 522–542.
37. Sokol, K., & Flach, P. (2023). Explainability fact sheets: A framework for systematic assessment of explainable approaches. Proceedings of the ACM Conference on Fairness, Accountability, and Transparency (FAccT), 56–67.
38. Verma, S., & Rubin, J. (2023). Fairness definitions explained. Proceedings of the International Workshop on Software Fairness (FairWare), 1–7.
39. Wachter, S., Mittelstadt, B., & Russell, C. (2023). Counterfactual explanations without opening the black box: Automated decisions and the GDPR. Harvard Journal of Law and Technology, 31(2), 841–887.
40. Zhao, H., Coston, A., Adel, T., & Gordon, G. J. (2023). Conditional learning of fair representations. Proceedings of the International Conference on Learning Representations (ICLR).
Additional Files
Published
Data Availability Statement
The study utilizes a publicly available dataset containing historical claim records for over 1.5 million auto insurance customers from a Taiwanese insurer, supplemented by open-access driver safety and speeding violation data. No new primary data were collected by the author.
