Using SMO, the predicted vlaes are very far from the facts.

Hi Mark,

I want to predict a numeric attribute using Support Vector Machine(SMO). Since the SVM can not predict Numeric values, I converted the numeric attributes to the Nominal.

The predicted values(0.0 ,0.0 ,14.0 ,19.0 ,11.0,......) are very far from the facts(1534 , 669.9 , 1156,.........). Why? what should I do to make it close to the facts?

Here is my code:

Filter filter = new NumericToNominal();

filter.setInputFormat(trainData);
Instances trainData1 = filter.useFilter(trainData, filter);

filter.setInputFormat(testData);
Instances testData1 = filter.useFilter(testData, filter);


trainData1.setClassIndex(trainData1.numAttributes()-2);
testData1.setClassIndex(testData1.numAttributes()-2);


SMO smo = new SMO();
String[] options1 = new String[1];
options1[0]= "-R";
smo.setOptions(options1);
smo.buildClassifier(trainData1);
for(int i = 1; i<= 20; i++)
{
double c = smo.classifyInstance(testData1.instance(i));

System.out.println(c);

}