k Nearest Neighbours: k = 1

Intro

Nearest Neighbour algorithms can be used for classification and regression.

Suppose we have the following training set:

We need to predict the label for a test object .

The algorithm will

  1. Search for the nearest training object to the test object
  2. Take the label of the nearest training object as the prediction for the label of the test object.

Binary Classification Problem

When our label space contains only two distinct values, this is binary classification.

The label space could be something like or , for instance. It doesn’t matter as long as we can distingish the two.

Euclidean Distance

In this example our label space is . Our training set consists of the following training objects:

Positive Objects:

Negative Objects:

Test Object:

We will use Euclidean Distance as our distance measure. We need to compute the Euclidean Distance between the test object and every training object.

To do this, we subtract vectors and compute the Euclidean norm (e.g. )

The Euclidean Distance between and the test object is computed like so:

Eventually, we have the Euclidean Distance for all training objects:

Now we make our prediction. We take the label of the object with the smallest Euclidean Distance. This is , which has the label .