Machine Learning for Climate Risk

Objective

This course explores how machine learning can be used to model, anticipate, and better manage climate-related risks in the insurance sector, with a specific focus on natural catastrophes such as floods, storms, droughts, and wildfires.

In a world where climate events are becoming more extreme and less predictable, actuaries and insurance professionals must rethink their traditional tools. This course provides a comprehensive understanding of current methodologies for climate risk modeling in insurance. It also aims to identify how and where machine learning can enhance these approaches — from data selection to the design of innovative models — while respecting operational and regulatory constraints (notably the CatNat regime in France).

Students will learn to work with a wide range of data — climatic, geospatial, and insurance-related — to enrich or reconstruct information on exposed assets. The goal is to turn these often fragmented and heterogeneous sources into meaningful features that power robust and explainable models tailored to today’s climate challenges.

Planning

By the end of the course, students will be able to:

• Analyze contemporary climate risks to ensure the viability of insurance systems and maintain accessible and sustainable coverage for policyholders.
• Understand the French regulatory framework for natural catastrophes (CatNat) and its implications for risk modeling.
• Identify, combine, and use climate, geospatial, and insurance data in practical modeling tasks.
• Apply supervised and unsupervised machine learning techniques to model:
  • Exposure to extreme weather events
  • Policyholder vulnerability
  • Climate-related claims
• Design and evaluate climate indicators relevant to insurance, including for parametric products.
• Build a full predictive modeling workflow: feature engineering, model training, validation, backtesting, interpretation, and operational integration.