Hey everyone, I'm working on different implementations of stacked generalization with Matlab, on a dataset which was reported to achieve higher performance scores than I'm actually getting. The issue is that with a stacking ensemble written by myself, that implemented by a postgraduate student, and finally one downloaded in a 3rd party toolbox get similarly low values. The dataset itself is highly unbalanced between 2 classes (99%/1% breakdown on average) and very noisy. Anyone had similar issues with weighted regression based meta-learners?
[link][comment]