the importance of parameters in machine learning with h2O

Dear R experts,

I used h2o packages to build training models to discriminate between targets and non-targets (methods including deeplearning, random forest, xgboost,general linear model) and the accuracy is generally above 80%. However, it is very difficult to interpret the importance of the variables. For instance, less targets live in a city A than non-targets. I would expect to see the relative/scaled importance is negative for the parameter "city A". However, the sign of the importance is positive. Does this mean that there is something wrong with my model building process or it is the problem of how h2o calculate parameter importance. Thanks for any suggestion in advance.

This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.

If you have a query related to it or one of the replies, start a new topic and refer back with a link.