Kernel basedsupportvectormachineviasemidefiniteprogramming:Applicationto medical diagnosis

Support vectormachine(SVM)isawellsoundlearningmethodandarobustclassificationprocedure. Choosing asuitablekernelfunctioninSVMiscrucialforobtaininggoodperformance;thedifficultyis how tochooseasuitabledatatransformationforthegivenproblem.Tothisend,multiplekernelmatrices, each ofthemcorrespondingtoagivensimilaritymeasure,canbelinearlycombined.Inthispaper,the optimal kernelmatrix,obtainedaslinearcombinationofknownkernelmatrices,isgeneratedusinga semidefinite programmingapproach.Asuitablemodelformulationassuresthattheobtainedkernelma- trix ispositivesemidefiniteandisoptimalwithrespecttothedatasetunderconsideration.Theproposed approach hasbeenappliedtosomeveryimportantmedicaldiagnosticdecisionmakingproblemsandthe results obtainedbycarryingoutpreliminarynumericalexperimentsdemonstratedtheeffectivenessof the proposedsolutionapproach.