I have to make comparison between 155 image feature vectors. Every feature vector has got 5 features. My image are divided in 10 classes. Unfortunately for using support vector machine i need at least 100 images for class, There is any alternative?
$\begingroup$
$\endgroup$
2
-
$\begingroup$ For automatic border creation, you can use alternatives such as NN's or something simpler, like logistic regression. I am assuming you have about 15 images per class? $\endgroup$– SpaceyMar 2, 2013 at 18:11
-
3$\begingroup$ Who said you must have at least 100 images per class for an SVM? It is usually good to have more data, but you can always try it with what you have and see what happens. $\endgroup$– DimaMar 4, 2013 at 16:45
Add a comment
|
2 Answers
$\begingroup$
$\endgroup$
4
Why not just stick with something simpler like k-nearest neighbors or (learning) vector quantization (PDF)?
-
$\begingroup$ Peter, I dont think KNNs will help here, because they help you classify data, whereas the OP is looking for ways to created the borders given the classifications already. (ie, he wants supervised learning, VS unsupervised learning). $\endgroup$– SpaceyMar 2, 2013 at 19:53
-
$\begingroup$ Mohammad, you may be right. I interpreted
i need at least 100 images for classto mean "I need at least 100 images for classification", which is certainly not the case for KNN or VQ. It seems to me the the OP is asking for other techniques to classify data. I gave two options. Perhaps the OP can clarify? $\endgroup$– Peter K. ♦Mar 3, 2013 at 0:55 -
-
$\begingroup$ @Thomas You can always go to the way back machine to find it. Added link to it. Please join the community and put some effort into keeping things updated! $\endgroup$– Peter K. ♦Dec 21, 2021 at 19:28
$\begingroup$
$\endgroup$
I think that the best option, probably, would be Random Forests or any other Boosting / Bagging method based on decision trees.
I would probably start with SK Learn Random Forests.
As more advanced trick I'd go after XGBoost.
