0
$\begingroup$

I am trying to train a Support Vector Machine (SVM) classifier to classify various items into 5 categories. I have trained two SVM classifiers, however, I am concerned that the accuracies and F1 scores do not change when some parameters are changed.

SVM Classifier 1 Parameters: c = 0.001, kernel = poly, degree = 5, coef0 = 2.5, tol = 0.001, gamma = auto

SVM Classifier 1 Results: Accuracy = 100% F1Score Weighted = 100%

If I were to change some of the parameters such as...

SVM Classifier 2 Parameters: c = 1.0 kernel = poly, degree = 1, coef0 = 0.0, tol = 0.001, gamma = auto

SVM Classifier 2 Results: Accuracy = 100% F1Score Weighted = 100%

Is it at all concerning that changing some of the parameters like C, degree, and coef0 do not change the overall results?

$\endgroup$
  • $\begingroup$ Does 100% accuracy make sense for any parameterization? It seems like you may just have a Very Easy problem that can be solved perfectly over a wide range of parameters. If you don't think the problem should be so easy to solve, you may have an issue in how you're computing performance metrics. $\endgroup$ – Nuclear Wang Jan 27 at 18:42
  • $\begingroup$ I do not believe the problem should solved this easily as the dataset I am working with is quite inter-twined and shows no good linear (or polynomial) separation. This is the reason I am concerned that a value of 100% accuracy is the result I am getting. What other items do you think I should look into or consider here? $\endgroup$ – Pythoner Jan 27 at 20:54
  • $\begingroup$ How would you detect overfitting in any supervised learning problem? $\endgroup$ – Sycorax says Reinstate Monica Jan 28 at 16:10
  • $\begingroup$ @SycoraxsaysReinstateMonica Depending on the SL algorithm, I would change the number of component of levels and see where the accuracy peaks, with the maximum being the best value to use. However, I have found that with SVMs this is not always the case. $\endgroup$ – Pythoner Jan 30 at 0:17
  • $\begingroup$ What is a component level and how it that relate to model selection? $\endgroup$ – Sycorax says Reinstate Monica Jan 30 at 0:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.