ENSAE Paris - École d'ingénieurs pour l'économie, la data science, la finance et l'actuariat

Artificial Intelligence in Insurance and Actuarial Studies

Objectif

We study during thoses lessons how to understand the main challenges of applying AI in insurance. We illustrate key concepts: ML modelling, industrializing models, implication of applying regulation such as GDPR, etc.  thought insurance examples. 

This class is essential for those wanted to evolve as (acturial) data scientists in the insurance industry. It helps to understand what are main challenges that we can solve and skills you migh want to develop after your degree.

Main objective of the course is to provide students a basic knowledge on what applying AI in insurance means. The course connects with actuarial notions (we remind definitions during the lessons for non actuarial students) and emphasize what it is needed to apply AI concepts in the industry (e.g. using Machine Learning to model a biometric event - spoiler, scikit-learn is not enough). We also share about what evolving as a data scientist in an insurance company looks like. 

Evaluation: written exam

Plan

The course is made of a mix of lessons and hands-on sessions.

During lessons we cover the topics below:

  • Reminder of the context of AI during the past 10 years and all challenges that came out (regulation, transparency, bias, cloud usage, etc.)
  •  Introduction of different Machine learning models and what they make them different than traditionnal statistical models
  •  Why we speak more about them today than 10 or 20 years ago ? what did change so that insurance is paying more attention?
  •  How to mitigate the risk of biais and mis-usage of machine learning? Illustration with some use cases.
  • What are the "new" skills we can expect from an actuarial data scientist?
  • Introduction to deep learning into insurance (NLP applications)
  •  What is model transparency? 

Hands-on sessions are in Python and usage of Git is essential (you might want to install it before sessions Git (git-scm.com) ) to illustrate some notions mentionned during the lessons.

Advised to people who have never used python to perform a quick training on it (will be provided during first session otherwise).

Références

nHastie, T., Tibshirani, R., Friedman, J. (2001) The Elements of Statistical Learning, Springer.
Chancel, Antoine and Bradier, Laura and Ly, Antoine and Ionescu, Razvan and Martin, Laurene and Sauce, Marguerite, (2022) Applying Machine Learning to Life Insurance: some knowledge sharing to master it 

Dimitri Delcaillau, Antoine Ly, Alize Papp, Franck Vermet, (2022) Applying Machine Learning to Life Insurance: some knowledge sharing to master it