The dataset we use is transformed from some real dataset

Numeric feature are nonlinear transformed

About 20% data become missing value

16 numeric features, 5 nominal features, 1 label

Our label is 'Weather'