Nearest Neighbor Classification Method Based on The Mutual Information Distance Measure

The k-nearest neighbor classification method predicts the class label of a query pattern based on its nearest neighbors. So which samples can be selected as the nearest neighbors of the query pattern and how to use these neighbor samples to predict the class label of the query pattern are two key problems in the nearest neighbor based method. Based on the definition of mutual information distance measure, a new classification method named mutual information distance measure nearest neighbor method is proposed in this paper. The efficiency of our method is validated and compared against other techniques using the real-world data, experimental results show that the new algorithm behaviors the excellent performance and robustness.