An Interpretable Deep Learning Model for General Insurance Pricing

2025-09-10

AI Summary

To address the trade-off between interpretability and predictive performance in general insurance pricing, this paper proposes an intrinsically interpretable Neural Additive Model (NAM). The model assigns dedicated subnetworks to each covariate and selected key interaction terms, and—novel in actuarial science—formally defines and achieves “full transparency”: predictions are fully traceable and parameters directly interpretable via enforced sparsity, monotonicity, and smoothness constraints. Integrating the structural interpretability of generalized linear models (GLMs) with the nonlinear modeling capacity of deep learning, the method incorporates an interaction selection mechanism and joint regularization. Experiments on synthetic data and real-world motor and household insurance datasets demonstrate that the proposed model significantly outperforms conventional GLMs, gradient-boosted trees, and standard NAMs—achieving both high predictive accuracy and end-to-end, fine-grained, mathematically verifiable interpretability.

Technology Category

• Machine Learning: Transparent, Interpretable, Explainable ML
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models
• Computer Vision: Interpretability, Explainability, and Transparency

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Uses of LLMs and GenAI for marketplace design, bidding, and strategic interactions

Abstract

This paper introduces the Actuarial Neural Additive Model, an inherently interpretable deep learning model for general insurance pricing that offers fully transparent and interpretable results while retaining the strong predictive power of neural networks. This model assigns a dedicated neural network (or subnetwork) to each individual covariate and pairwise interaction term to independently learn its impact on the modeled output while implementing various architectural constraints to allow for essential interpretability (e.g. sparsity) and practical requirements (e.g. smoothness, monotonicity) in insurance applications. The development of our model is grounded in a solid foundation, where we establish a concrete definition of interpretability within the insurance context, complemented by a rigorous mathematical framework. Comparisons in terms of prediction accuracy are made with traditional actuarial and state-of-the-art machine learning methods using both synthetic and real insurance datasets. The results show that the proposed model outperforms other methods in most cases while offering complete transparency in its internal logic, underscoring the strong interpretability and predictive capability.

Problem

• Creating interpretable deep learning for insurance pricing transparency
• Balancing neural network predictive power with actuarial interpretability requirements
• Addressing need for transparent models while maintaining strong prediction accuracy

Innovation

• Neural additive model with dedicated subnetworks per covariate
• Architectural constraints for interpretability like sparsity and smoothness
• Combines neural network predictive power with full transparency