Title: A framework for multidimensional learning using multilabel ranking
Authors: S. Shuaib Ahmed; B. Purna Chandra Rao; T. Jayakumar
Addresses: Nondestructive Evaluation Division, Indira Gandhi Center for Atomic Research, Kalpakkam – 603 102, TN, India ' Nondestructive Evaluation Division, Indira Gandhi Center for Atomic Research, Kalpakkam – 603 102, TN, India ' Nondestructive Evaluation Division, Indira Gandhi Center for Atomic Research, Kalpakkam – 603 102, TN, India
Abstract: Multidimensional learning problem which is a more general form of supervised machine learning dealing with learning a function that maps a vector of input features to a vector of class labels is discussed in this paper. The study focuses on a framework which uses multilabel ranking with an emphasis on solving multidimensional learning problems. Performance of the framework is evaluated using mean accuracy, global accuracy and entropy of accuracy. Studies using real world benchmark dataset have shown that all the assumptions made in building the framework are valid. In addition to this, studies are carried out on images from eddy current non-destructive testing of stainless steel plate having subsurface defects clearly confirm that the proposed framework is well suitable for learning multidimensional problems and is superior to uni-dimensional learning algorithms.
Keywords: multidimensional learning; multilabel ranking; entropy of accuracy; mean accuracy; global accuracy; eddy current testing; machine learning; supervised learning; class labels; vectors; input features; nondestructive testing; NDT; stainless steel plate; subsurface defects.
International Journal of Advanced Intelligence Paradigms, 2013 Vol.5 No.4, pp.299 - 318
Received: 01 Feb 2013
Accepted: 22 Mar 2013
Published online: 30 Jul 2014 *