| Autor: |
GAWLOWSKI, KAROL, CONDON, JOHN, HARRINGTON, JACK, RUFFINI, DAVIDE |
| Předmět: |
|
| Zdroj: |
Actuary; Mar2024, p34-37, 4p |
| Abstrakt: |
This article discusses two new modeling technologies, the combined actuarial neural network (CANN) and the LocalGLMnet, which aim to improve predictive accuracy in the field of pricing. The CANN combines a neural network with a benchmark generalized linear model (GLM) to enhance the model's predictive power. The LocalGLMnet uses a neural network to predict GLM coefficients for each individual record, allowing for greater accuracy by varying the coefficients. The article presents results from testing these models on real-world data and concludes that these newer architectures have more predictive power than traditional GLM models, although sacrifices in transparency and explainability may be made. [Extracted from the article] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
