grid search and cross validation

Make sure you have a read of A practical guide to SVM classification, available from the LibSVM website.

If you use wekas .arff data format, try using this code (import jars from weka and libSVM):

import java.util.Random;
import libsvm.svm;
import libsvm.svm_print_interface;
import weka.classifiers.Classifier;
import weka.classifiers.Evaluation;
import weka.classifiers.functions.LibSVM;
import weka.core.Instances;
import weka.core.converters.ConverterUtils.DataSource;

public class SVMtester {

    private static svm_print_interface svm_print_null = new svm_print_interface()
    {
            public void print(String s) {}
    };

    public static void main(String[] args) throws Exception {

        String fileLocation =  "C:\\Documents and Settings\\Data\\";
        String fileName = "datafile.arff";

        svm.svm_set_print_string_function(svm_print_null);

        Classifier classifier = null;
        DataSource source = null;
        Instances dataInstances = null;
        Evaluation evaluation = null;

        int sizeOfSearch = 11;  //or...

        double[][] resultsArray = new double[sizeOfSearch][sizeOfSearch];

        int kernel = 2;//see weka for parameters

        //this is the power of the cost parameter 2^c
        for(int c =  0; c < sizeOfSearch; c++){
            //this is the power of the gamma parameter 2^g
            for(int g = 0; g < sizeOfSearch; g++){

                int cVal = 2*c - 5;  //look at libsvm guide, change if you want
                int gVal = 2*g - 5;
                double cost = Math.pow(2, cVal);
                double gamma = Math.pow(2, gVal);

                classifier = new LibSVM();

                String[] options = {"-S", "0", "-K", kernel+"", "-D", "3", "-G", gamma+"", "-R", "0.0",
                        "-W", "1.0 2.0","-N", "0.5", "-M", "40.0", "-C", cost+"", 
                        "-E", "0.01", "-P", "0.1"};
                classifier.setOptions(options);

                source = new DataSource(fileLocation + fileName);
                dataInstances = source.getDataSet();
                dataInstances.setClassIndex(dataInstances.numAttributes()-1);
                classifier.buildClassifier(dataInstances);              
                evaluation = new Evaluation(dataInstances);
                evaluation.crossValidateModel(classifier, dataInstances, 10, new Random(1));

                double correct = roundToDecimals(evaluation.pctCorrect(),2);

                resultsArray[c][g] = correct;
                System.out.println("cost: " + cost + "\tgamma: " + gamma + "\t" + correct);

                classifier = null;
                source = null;
                dataInstances = null;
                evaluation = null;

            }
        }
        System.out.println();
        print(resultsArray);
    }

    public static void print(double[][]in){
        for(int i = 0; i < in.length; i++){
            for(int j = 0; j < in[0].length; j++){
                System.out.print(in[i][j] + "\t");
            }
            System.out.println();
        }
    }

    public static double roundToDecimals(double d, int c){
        int temp = (int)((d*Math.pow(10, c)));
        return (((double)temp)/Math.pow(10,c));
    }
}

(I'm sure the code can be nicer, but it'll work).

Thanks Amair.