On support vector machines for data classification