1
$\begingroup$

I have 150 pictures that represent archeological signs and 5 categories to which they belong. These pictures have features like circularity, roughness and elongation that are expressed as continuous values, and other values like number of end points, number of contours, and angles between segments.

I have read that SVM needs at least 100 images for every category to work well, nearest neighbor use Euclidean distance metric that does not consider all features in different ways, mix everything! Alternatives?

What would be the best classifier?

$\endgroup$
  • $\begingroup$ Hi Pitone, have you tried running it and doing cross validation? You could compare the misclassification rate to other classifiers such as random forest. (I'm not sure the 100 image rule is a rule, the number you need will be dependent on how messy the classification boundaries are) $\endgroup$ – conjectures May 4 '13 at 7:48
  • $\begingroup$ I tried GridSearchCV and cross validation, what score (precision, recall) I get at least because it is reliable? $\endgroup$ – Pitone Jun 8 '13 at 0:47
  • 1
    $\begingroup$ @Learner, I believe "neighbour" is the British spelling of neighbor. Note that some CV users are British, or otherwise use British spelling, & tend to take offense when British English spelling is treated as incorrect & changed to American English spelling. $\endgroup$ – gung Feb 15 '15 at 15:17
  • $\begingroup$ @gung: My Apologies for that & I will keep your advice in my mind. :) $\endgroup$ – Learner Feb 15 '15 at 15:22

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.