Similarity Graph Neighborhoods for Enhanced Supervised Classification

We consider transformations to enhance classification accuracy that are applied during the training phase of a supervised classifier with prelabeled data. We consider similarity graph neighborhoods (SGN) in the feature subspace of the training data to obtain a transformed dataset by determining displacements for each entity. Our SGN classifier is a supervised learning scheme that is trained on these transformed data; the separating boundary obtained is thereafter used for future classification. We discuss improvements in classification accuracy for an artificial dataset and present empirical results for Linear Discriminant (LD) classifier, Support Vector Machine (SVM), SGN-LD, and SGN-SVM for 6 well-known datasets from the University of California at Irvine (UCI) repository [1]. Our results indicate that on average SGN-LD improves accuracy by 5.0% and SGN-SVM improves it by 4.52%.