We build a weighted random forests classifier to identify sinkholes.

We select depression geometry and natural/human factors as predictive variables.

Higher weight for the sinkhole class improved accuracy in identifying true sinkholes.

Including variables besides shape and depth of depressions improved the classifier.