Whereas advanced Machine Learning (ML) techniques (e.g. random forest or neural networks) usually have a better predictive power than statistical techniques (e.g. GLM), their main drawback is that they are black-box and their results are difficult to understand or interpret.
There are basically 2 strategies to use ML techniques in predictive modelling:
1. Replacing traditional models (e.g. GLM) by ML models
2. Combining the pros of traditional and ML models to improve predictive modelling
The goals of this presentation are therefore to
- Briefly remind some useful machine learning techniques and explain why it is difficult to interpret their results
- Present several techniques that have been developed in order to better understand the results of machine learning techniques
- Explain how these interpretation techniques can be used to implement the 2 strategies presented above and improve predictive modelling
NB: This presentation was given as part of the IACA webinar “Explainable machine learning for actuaries”.
There are basically 2 strategies to use ML techniques in predictive modelling:
1. Replacing traditional models (e.g. GLM) by ML models
2. Combining the pros of traditional and ML models to improve predictive modelling
The goals of this presentation are therefore to
- Briefly remind some useful machine learning techniques and explain why it is difficult to interpret their results
- Present several techniques that have been developed in order to better understand the results of machine learning techniques
- Explain how these interpretation techniques can be used to implement the 2 strategies presented above and improve predictive modelling
NB: This presentation was given as part of the IACA webinar “Explainable machine learning for actuaries”.