kNN-Classifier

kNN classifier built in MATLAB. It is tested on five datasets:

• Iris
• Wine
• Banknote Authentication
• Ionosphere
• Magic Gamma Telescope

Datasets are taken from UCI Machine Learning Repository.

Structure of repository

Repository contains five folders for each dataset. Each folder further contains two files:

• Main index file 'main.m'
• KNN Classifier function 'knnclassifier.m'

Accuracy plots are also included in the folder of each dataset.

Implementation

First data is stored and divided into y​equal parts (y​­​fold). One part is declared as test data and rest is training data. This completes the training phase. During test phase, a test sample is picked and all the training samples are sorted according to normal or (weighted) euclidean distance from test sample. For first k​data points (in sorted list) polling is done. The class with maximum frequency is allotted to test data sample. Same procedure is repeated for all the test data points. For a particular dataset, k is varied from 1 to 5 and y is varied from 2 to 5.

Tie break: It may happen when k is even. Two classes may have same frequency during polling. In this case, sum of distances for both the classes is calculated. Class with minimum sum is allotted to test data sample.