An Interpretable Deep Learning Model for General Insurance Pricing
提出一种内在可解释的深度学习模型，为每个协变量和交互项分配独立子网络，在保持预测能力的同时满足保险定价的平滑性、单调性等实际要求，并在合成和真实数据集上优于传统方法。
The rapid advancement of machine learning has provided an opportunity to transform the modeling techniques in actuarial analytics. Novel machine learning methods, especially deep learning, have demonstrated versatile modeling capability and superior predictive performance compared to traditional actuarial approaches such as Generalized Linear Models. However, the widespread adoption of deep learning techniques in the insurance industry is often hindered by the lack of model interpretability, as the intricacies of their inner workings remain obscured behind the complex model architecture. This lack of interpretability is further complicated by the absence of a generally accepted definition of what an interpretable model is. There are also various practical requirements, such as smoothness and monotonicity, that a pricing model in general insurance should possess in addition to being interpretable. 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.