1
$\begingroup$

I'm trying to implement a multiclass classification on a 25 features/1300 samples dataset using LinearSVC from sklearn. Unfortunately, my results both on the training and the test sets are very poor (60%). I've tried to optimize the combination of parameters using also GridsearchCV but the results are still the same.

My question is: is there something that I can try in order to improve the performance of the classifier? PCA and standardization? Is it possible that is just this classifier that is not suited to work on the kind of dataset I have? Why could that be?

$\endgroup$

2 Answers 2

0
$\begingroup$

Your data could be not separable in linear space. Try using a non-linear method like kernel SVC or RandomForests. Be careful to avoid overfitting!

$\endgroup$
0
$\begingroup$

I think you could try both to do some feature engineering and to use other classifiers. Also a good tuning of the parameters may help, but based on my experience this has a lower impact than a good features selection and the choice of the classifier. Scikit-learn.org offers wide spectrum of classifiers. Also, always do cross-validation.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.