A Hyper-Solution Framework for SVM Classification: Application for Predicting Destabilizations in Chronic Heart Failure Patients
Antonio Candelieri*, Domenico Conforti*
Identifiers and Pagination:Year: 2010
First Page: 136
Last Page: 140
Publisher Id: TOMINFOJ-4-136
Article History:Received Date: 5/12/2009
Revision Received Date: 5/3/2010
Acceptance Date: 15/3/2010
Electronic publication date: 27/7/2010
Collection year: 2010
open-access license: This is an open access article licensed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0/) which permits unrestricted, non-commercial use, distribution and reproduction in any medium, provided the work is properly cited.
Support Vector Machines (SVMs) represent a powerful learning paradigm able to provide accurate and reliable decision functions in several application fields. In particular, they are really attractive for application in medical domain, where often a lack of knowledge exists. Kernel trick, on which SVMs are based, allows to map non-linearly separable data into potentially linearly separable one, according to the kernel function and its internal parameters value. During recent years non-parametric approaches have also been proposed for learning the most appropriate kernel, such as linear combination of basic kernels. Thus, SVMs classifiers may have several parameters to be tuned and their optimal values are usually difficult to be identified a-priori. Furthermore, combining different classifiers may reduce risk to perform errors on new unseen data. For such reasons, we present an hyper-solution framework for SVM classification, based on meta-heuristics, that searches for the most reliable hyper-classifier (SVM with a basic kernel, SVM with a combination of kernel, and ensemble of SVMs), and for its optimal configuration. We have applied the proposed framework on a critical and quite complex issue for the management of Chronic Heart Failure patient: the early detection of decompensation conditions. In fact, predicting new destabilizations in advance may reduce the burden of heart failure on the healthcare systems while improving quality of life of affected patients. Promising reliability has been obtained on 10-fold cross validation, proving our approach to be efficient and effective for an high-level analysis of clinical data.