L4_Intro: Using past to predict future

[ex-man]

In this lesson, the data will be analysed further to find a GLM model, that is consistent over time and represents a good fit for some years of data. Concepts of validation and other diagnostics may be introduced.

[ex-man]

• [ ] data prep: Exposure prep, Claims prep, merging, maybe some cleaning
• [ ] one way: indentify target, why, (Loss / Exposure [Time of Policy]), then some avgs per predictors
• [ ] decision tree: which predicstor is more significant? / info for grouping to glm?

[ex-man]

found this reasonable visual of glm... hope it helps showing what it is... http://www.brainvoyager.com/bvqx/doc/UsersGuide/StatisticalAnalysis/TheGeneralLinearModel.html

[cvoky]

Found little bit strange thing or I do not understand data well.

What is the right level of policy data to ensure correct merging with claims data? I was expecting to have unique observetion per NrPolicy + NrObject.

Found duplicates in identifiers, see below:

Please explain. For now excluding those duplicates.

---

Another strange thing to me is why one claim can be linked to two different policy no. ?

Please explain. For now excluding those duplicates.

[ex-man]

Both can happen. 2nd one if there are 2 policyholders that hit each other in one accident. 1st, if there are more insureds on one policy (but likely an error in data, unless this si after join, in which case there were 2 claims, which is fine)

[cvoky]

Thanks! Make sense. For now I would suggest to not include those cases, so we can more focus to the core of the lesson.

What do you think?

[ex-man]

Sounds good. Feel free to remove.