BEYOND GLMs – AN INTRODUCTION TO ADVANCED PREDICTIVE ANALYTICS

Jean-Francois Greeff

Presentation

17 October 2017 | 13:15 – 14:15 and 14:45 – 15:45 | Committee 2

Relevant practice area(s):Data Analytics

Suggested audience knowledge level: Foundational and Intermediate

Abstract:

This two-part workshop will serve as an introduction to advanced predictive analytics for the actuary who is familiar with Generalized Linear Models. During the workshops, attendees will discuss some of the problems faced by modellers when working with GLMs and be introduced to methods used to address these problems. The workshop will be interactive and having a laptop with R is recommended.

Part 1 will be dedicated to discussing Tree-based models. Topics will include: Classification and Regression Trees; pros and cons of trees; bagging, and random forests. Time permitting, we will introduce the concept of boosting and gradient boosted machines.

Part 2 will be dedicated to discussing Regularized Regression. Topics will include: “Wide” data and the problem of variable selection; the curse of dimensionality; penalized regression; ridge, Lasso, and elastic net lasso regression; and using cross-validation to estimate tuning parameters.