A Performance Analysis Of The Svm Technique For Different Kernels

This paper reports the results of a series of experiments on the machine learning technique known as SVM, for the comparison of five different kernels. First, we compare and analyze their classification performance on databases with missing values. Then, we run a similar experiment with noisy values databases. Finally, we analyze the results from adding irrelevant features to the databases. The SVM technique turned out to be very robust in the three created environments, and the polynomial kernel showed the best classification results.