?embed=1 for iframe mode kNN (k-Nearest Neighbours) is a simple classification algorithm. It predicts the class of a new data point by looking at the k closest labelled points and taking a majority vote.
What does kNN use to classify new data?
Our dataset has two clear clusters of points. Each point is labelled as either Class A or Class B, forming distinct groups in the space.
What pattern do you see in the dataset?
With k = 1, the algorithm looks at only the single nearest neighbour. The new point gets classified as whatever class that nearest neighbour belongs to.
How many neighbours does k = 1 consider?
With k = 3, the algorithm examines the three nearest neighbours and uses majority voting. If two are Class A and one is Class B, the prediction is Class A.
How does k = 3 make its prediction?
With k = 7, we consider seven nearest neighbours. This larger k value makes the classification more stable and less sensitive to individual outlier points.
What advantage does a larger k provide?
The decision boundary shows which regions would be classified as A or B. With k = 1, the boundary is very jagged because it reacts to every single nearby point.
Why is the k = 1 boundary jagged?
With k = 7, the boundary becomes much smoother. Instead of following every local variation, it captures the general separation between the two clusters.
What does a smooth boundary indicate?
Side-by-side comparison shows the key trade-off: small k is sensitive to local details (can overfit), while large k focuses on broader patterns (more generalizable). The transition zone between clusters shows this most clearly.
In the ambiguous middle region, which k is more stable?