How do I judge whether the dataset is unbalanced (is it when the minority class inferior of 15%) could us use the test CR after balancing the data?
Asked
Active
Viewed 34 times
0
-
I find it somewhat unclear what you're asking. In principle, unbalanced means there is less (or more) than a $\frac{1}{k}$ ratio of observations in each class, where $k$ is the number of classes. How 'bad' unbalanced data is, depends wholly on the algorithms you use to model your data. – Niels Wouda Jun 21 '17 at 10:52
-
@N.Wouda Thank you for your feedback, okey if I get it well it is the ratio 1/k which gives me an idea about whether my data is unbalanced or not ( am a bit confused since I found on some tuto that if the rate of minority class is less than 15% of the whole observation we talk about unbalanced data.. ) – mirette Jun 21 '17 at 11:30
-
related to what you have said suppose that we have 81 is the minority class instances ( the number of observations) and 225 the number of the other class (306 is the toal data set here since the minority class' 0.26...< 0.5 = (1/k = 1/2 ) according to what you said ---> in this case we have unbalanced data is not it ! am using the LSSVM ( least square support vector machine) a version of the SVM algorithm – mirette Jun 21 '17 at 11:31