A Tree-like Representation of the Training Set for Classification Time Reduction

J. Lobato, J.S. Sánchez, and J.M. Sotoca (Spain)


Pattern Recognition; Classification; Neighborhood; Tree Design Algorithm; Space Partition.


In this paper, we introduce a new algorithm to efficiently search for neighbors of a test sample. The main advantage of this technique over other search algorithms refers to the fact that the one proposed here can be applied to any kind of neighborhood. Thus the algorithm results in a tree-like data structure adapted to the specific type of neighborhood to be further employed by a distance-based classifier. Experiments over several synthetic data sets illustrate the benefits of using the method introduced in this paper in terms of classification time.

Important Links:

Go Back