6.00pm - 6.30pm: Tea/Coffee Reception
6.30pm - 8.00pm: Meeting
When asked for their favourite machine learning technique, many data scientists will name the LASSO which simultaneously:
- Selects a model and shrinks parameters
- Produces a readily interpretable model
- Is applicable to a wide range of problems.
In this talk we will first give an overview of the LASSO before considering its application to a particular problem in general insurance – the estimation of loss reserves. However, we note that similar techniques can and have been used for other problems in all areas of insurance and beyond, so will aim to make this talk accessible to non-GI actuaries.
We will discuss some issues around feature engineering which are crucial to the good performance of the lasso. Finally we will demonstrate the use of the LASSO on synthetic data sets with known features and on a real data set, and will show how the LASSO can be used for automated modelling of highly complex data sets.
A paper discussing the contents of this presentation is available at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3241906.