Pick a digit and drag the slider to remove its samples. Watch the imbalance ratio grow — and watch the lazy-model warning appear.
Loading...
Loading...
Loading...
Think Deeper
Try this:
In the Explore tab, remove 95% of digit 5. What would happen if a model just guessed the most common digit every time?
With ~9 samples of digit 5 vs ~1,600 of everything else, a model that never predicts 5 gets ~99.5% accuracy. This sets up the accuracy trap in Step 6.
Cybersecurity tie-in: In real-world security data, attacks are the rare class.
99.9% of traffic is normal. A model that says "normal" every time gets 99.9% accuracy — but
catches zero attacks. This is why class balance matters.