KNN++: An Enhanced K-Nearest Neighbor Approach for Classifying Data with Heterogeneous Views

Abstract

In this paper, we proposed an enhanced KNN approach, which is denoted as KNN++, for classifying complex data with heterogeneous views. Any type of view can be utilized when applying the KNN++ method, as long as a distance function can be defined on that view. The KNN++ includes an integral learning component that learns the weight of each view. Furthermore, the KNN++ method factors in not only the training data, but also the unknown instance itself when assessing the importance of different views in classifying the unknown instance. Experimental results on predicting SPY daily open price demonstrates the effectiveness of this method in classification. The time complexity of the KNN++ method is linear to the size of the training dataset.