Forensic analysis of whiskey

K-nearest neighbors analysis

• Predicts group membership of sample based on group membership of its K-nearest neighbors.
  1. Calculate interpoint distances for all samples.
  2. Look at the target samples' K-nearest neighbors (where K typically takes the value of 1 and 3-9), tally their group membership.
  3. The group receiving the most votes for each sample and for each value of K is the predicted group for that sample and value of K.

• In the whiskey example, each case was considered a target sample, to test if the natural groupings hold together.

Number of misclassified samples (of 68)
K	new and aged premium (34)	"counterfeits" (34)
1	0	1
3	0	1
5	0	1
7	0	1
9	0	2

• A natural separation exists between the samples.

• The criteria of a quick, simple and easy analysis still not met.

Previous slide  |   Next slide  |   Table of Contents  |   Jump to presentation text