Advertisement
Linear models = straight boundary. k-NN = local, can be very irregular.
What you're seeing
Decision boundary: where the classifier flips its prediction. Linear models always produce a hyperplane. k-NN follows local majority — produces complex, non-linear boundaries.
Trade-off: linear models are simple/regularized/fast; k-NN/trees/NNs adapt to non-linear patterns at the cost of more data and risk of overfit.
★ KEY TAKEAWAY
Linear models: straight boundary. k-NN: local, complex. Same data, very different fits.
▶ WHAT TO TRY
- Switch between Linear and k-NN.
- Click Reset data for new clusters.
- Linear underfits non-linear data; k-NN can overfit.