Multi Class classification approach for files in 20 different categories

[Abhinav Katiyar]

Hi,

I m learning and exploring more on Accord.NET and was comparing with R. I have following code in R

"svm_model <- svm(Doccategory~.,data = trainDataSetAll[trainInd,], probability = TRUE)SVM-Type: C-classification

SVM-Kernel: radial

cost: 1

gamma: 0.0005157298"

I tried the following code in Accord

 IKernel kernel = new Gaussian();            

            var machine = new MulticlassSupportVectorMachine(totalColumns, kernel, 12);            

            var teacher = new MulticlassSupportVectorLearning(machine, inputs, outputs);

            teacher.Algorithm = (svm, classInputs, classOutputs, i, j) => new SequentialMinimalOptimization(svm, classInputs, classOutputs);

            double error = teacher.Run();

Issue:

The accuracy i achieved was 61% while in R it was 86%, I have tried various Kernels(Polynomial,Taylor Gaussian) but accuracy is down. Can you suggest any equivalent approach in Accord for the R code pasted above ?

I have to classify certain files in 20 different categories.

Thanks,

Abhinav