A new mathematical approaches to large scale classification problems

Classification problem is called that the problem is to estimate class of new instances by using models that is constructed on the data sets which are known class labels.Researchers from different disciplines study on new approaches to solve this problems encountered in many areas. Hence new approaches and solution methods are presented in this field.In this thesis, a new mathmematical programming approach based on polyhedral conic functions is presented to solve large scale classification problems. In order to solve problems effectively and quickly with this approach; k-means and robust linear programming are used. Frequently used large scale test problems from the literature are solved either proposed approach or commonly used methods with proven success in the field. Obtained results are shown that new approach is clearly superior than the others.