1
$\begingroup$

I need some advice on methodology.

I need to predict a numeric value (claim amount) being as good as possible on high values corresponding to rare events (corporal damage, technological disaster...).

Any suggestion on methodology would be appreciated. Thanks.

$\endgroup$
2
  • 1
    $\begingroup$ ...being as good as possible on high values ... Could you clarify on this phrase please. $\endgroup$
    – naive
    Apr 27, 2019 at 15:45
  • $\begingroup$ My purpose is to predict a rare event predicting the cost of that event as I know this event has high cost ( and I know every high cost events are rare ). I tolerate predicting a small cost event as a high cost event but it is less acceptable to predict a high cost event as a low cost event : false negative cases must be avoided (but I am on a regression model not really a classification one). The predictions should not just be good on the main most occuring events. $\endgroup$ Apr 28, 2019 at 14:49

1 Answer 1

1
$\begingroup$

My suggestion:

There is a technique called Quantile Regression, which allows you to determine the highest (or lowest) values of a variable. This is valid only if you have regressors (exogenous variables).

$\endgroup$
1
  • $\begingroup$ Looks like a good direction to follow. $\endgroup$ Apr 28, 2019 at 17:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.