Algorithm FRiS-TDR for Generalized Classification of the Labeled, Semi-labeled and Unlabeled Datasets

Abstract

The problem of generalized classification combines three well-known problems of machine learning: classification (supervised learning), clustering (unsupervised learning), and semi-supervised learning. These problems differ from each other based on the ratio of labeled and unlabeled objects in a training dataset. In the classification problem all the objects are labeled, and in the clustering problem all the objects are unlabeled. Semi-supervised learning makes use of both labeled and unlabeled objects for training—typically a small amount of labeled objects with a large amount of unlabeled objects. Usually these problems are examined separately and special algorithms are developed for solving each of them. Algorithm FRiS-taxonomy decision rule based on function of rival similarity examines these three problems as special cases of the generalized classification problem and solves all of them. This algorithm automatically determines the number of clusters and finds effective decision rules independently of the ratio of labeled and unlabeled samples in datasets.