The data could be classified correctly by method like LogisticRegression, NearestNeigbors.

The dataset is quiet simple, dataset wdx

This is not the first time I encounted with this method which across many versions.

Tested on 13.3 on windows 10.

As you can see, ListPlot,SetAlphaChannel many interesting symbols appear in the error information. This error information seems appear in the final report stage, not training stage.

Some sliced dataset is OK to train without error.

Split problem dataset into 2 is OK.

A random dataset is OK, so the problem is related with some certain data.

dataset=Thread[RandomReal[{-1,1},{2000,40}]->RandomInteger[40,2000]];