Centro de Documentacion de Fundación MAPFRE - Accelerating the computation of shapley effects for datasets with many observations

MAP20260002972

Accelerating the computation of shapley effects for datasets with many observations / Giovanni Rabitti and George Tzougas

Sumario: The document presents a strategy to accelerate the computation of Shapley effects, a sensitivity-analysis method used to identify the importance of risk factors in actuarial models. The traditional procedure becomes computationally expensive when dealing with large datasets. The authors propose reducing the sample size using techniques such as Latin Hypercube Sampling, Conditional Latin Hypercube Sampling, and Hierarchical k-means, selecting representative observations while preserving calculation accuracy. They apply this methodology to the well-known French automobile claim-frequency dataset, demonstrating drastic reductions in computation time with minimal loss of precision. The study concludes that this approach enables efficient estimation of Shapley effects even in big-data contexts, providing a relevant advancement for actuarial modeling and insurance risk analysis

En: European Actuarial Journal. - Cham, Switzerland : Springer Nature Switzerland AG, 2021-2022. - 11/08/2025 Volume 15 - Number 2 - August 2025 , p. 885 - 898

1. Big data . 2. Modelos actuariales . 3. Clusters . 4. Siniestralidad . 5. Machine learning . I. Tzougas, George . II. Springer Nature . III. Título.