The Novel k Nearest Neighbor Algorithm

In the field of Text Classification/Categorization, the k Nearest Neighbor algorithm (kNN) has been to date one of the oldest and most popular methods. It has been experimented upon, implemented and tested by many researchers all over the world. There have been variations in the implementation of this algorithm and I have in this paper done the same. As the name suggests the method is dependent on the parameter 'k' which can drastically change the output as we vary its values. When the training set contains classes of unequal sizes, the test data is likely to get classified to a class which has more samples than the actual class it belongs to, if that actual class has less number of samples. In the proposed method, I have added a small variation to the classic kNN and have named this method 'The Novel k Nearest Neighbor Algorithm'. The parameter k in this method depends on the size of the smallest class sample.