I use "LIBSVM" library for classification.
If my first class is the same as the second class, for accuracy I will get 50% or 0%.
What does it mean to get 30% or 20% of precision?
$\begingroup$
$\endgroup$
7
-
$\begingroup$ I still fail to understand the experiment you are doing. What do you mean by "the first class is the same as the second class"? If they are same, they should be in one class? From your comment to Harald's answer, ("I ask two subjects to the same movement"), I guess that you have 2 samples from one class. You should at least ask those 2 subjects to make a different movement, then you will have 4 samples from 2 classes. And I don't know what type of features you are using but a dataset of 2 samples is definitely very small. $\endgroup$– jeffFeb 17, 2016 at 15:18
-
$\begingroup$ it is our experience, we want to see the resulat between 2 subjects make the same mouvement $\endgroup$– ensbergFeb 17, 2016 at 15:24
-
$\begingroup$ So you are trying to identify people based on their movement patterns? If so, i guess you are right, your classes are the people. In this case, reporting the classification accuracy can make more sense though. Are you sure you need to measure the precision, but not the classification accuracy? $\endgroup$– jeffFeb 17, 2016 at 16:28
-
$\begingroup$ please can you see me what is the difference between, precision and classification accuracy $\endgroup$– ensbergFeb 18, 2016 at 1:46
-
$\begingroup$ if the problem was image retrieval with classes car, human and boat, and the test image was infact a car image, then your precision would be (the number of car images you returned)/(number of car images in your database). there is also another measure called recall, both are explained here. $\endgroup$– jeffFeb 18, 2016 at 2:37
|
Show 2 more comments
1 Answer
$\begingroup$
$\endgroup$
14
A SVM is a binary classifier. That means, you can distinguish only 2 classes.
If both classes are the same, classification doesn't make sense.
If you have more then 2 classes you need to use different classification methods, such as k-means or neural networks. However, multi-class classification with SVM can still be done by reducing the multi-class decision problem to multiple binary classifications.
-
$\begingroup$ my experience is, I take two set of data generated by two subject that do the same thing, I get almost the Same data, $\endgroup$– ensbergFeb 17, 2016 at 12:31
-
$\begingroup$ Can you explain your question a little more detailed? $\endgroup$ Feb 17, 2016 at 12:45
-
$\begingroup$ I classify the movements, I ask two subjects to the same movement. I get almost the same data, what does it mean to get 50% of accuracy $\endgroup$– ensbergFeb 17, 2016 at 13:00
-
$\begingroup$ Sorry, I can't help you if you explain yourself this scarce. $\endgroup$ Feb 17, 2016 at 13:02
-
$\begingroup$ I want just to understand the meaning of obtaining 50% of preccision with SVM $\endgroup$– ensbergFeb 17, 2016 at 13:04