A Novel Partial-Memory Learning Algorithm Based on Grey Relational Structure

Abstract

In instance-based learning, the storage of instances must increase along with the number of training instances. In addition, it usually takes too much time to classify an unseen instance because all training instances have to be considered in determining the ‘nearness’ between instances. This paper proposes a novel partial-memory learning method based on the grey relational structure. That is, only some of the training instances are adopted for classification. The relationships among instances are first determined according to the grey relational structure. In this relational structure, the inward edges of each training instance, indicating how many times each instance is used as the nearest neighbor or neighbors in determining the class labels of other instances, can be found. This approach excludes the training instances with no or few inward edges for learning. By using the proposed approach, new instances can be classified with a few training instances. Five datasets are used for demonstrating the performance of the proposed approach. Experimental results indicate that the classification accuracy can be maintained when most of the training instances are pruned prior to learning. Meanwhile, the number of remained training instances is comparable to that of other existing pruning techniques.