Name: Feature Selection Lab
Author: MySimulator

Dataset

Noise features to add: 20 Training samples: 40

Only 2 of the (2 + noise) features actually determine the label. The rest are pure random noise — yet with few samples, a model can still find spurious patterns in them.

Accuracy on held-out test set

Using all features—

Using only top-2 selected—

Test accuracy vs. number of top-ranked features kept