Max-min separability: incremental approach and application to supervised data classification

Adil M Bagirov(a.bagirovballarat.edu.au)
Dean Webb(d.webbballarat.edu.au)
Bulent Karasozen(bulentmetu.edu.tr)

Abstract: A new algorithm for the computation of a piecewise linear function separating two finite point sets in $n$-dimensional space is developed and the algorithm is applied to solve supervised data classification problems. The algorithm computes hyperplanes incrementally and it finds as many hyperplanes as necessary to separate two sets with respect to some tolerance. An error function is formulated and an algorithm for its minimization is discussed. We present results of numerical experiments using several UCI test data sets and compare the proposed algorithm with two support vector machine solvers: LIBSVM and SVM$\_$light.

Keywords: Separability, nonconvex optimization,nonsmooth optimization, supervised data classification.

Category 1: Global Optimization (Applications )

Category 2: Convex and Nonsmooth Optimization (Nonsmooth Optimization )

Citation: http://www3.iam.metu.edu.tr/iam/images/0/03/Preprint76.pdf Preprint Series of Institute of Applied Matheamtics, Middle Eat Technical University, Ankara-Turkey

Download: [PDF]

Entry Submitted: 06/25/2007
Entry Accepted: 06/29/2007
Entry Last Modified: 06/25/2007

Modify/Update this entry