kindly tell me if i am using a dataset which have to produced binary class attributes like Student Result "Pass" or :Fail". i have data which have 80% pass students and 20% fail. in reality it is true because same ratio is observed in real life however due to problem of intention the classifier towards majority class here i think needs to be it balanced. the question is that in this case 50/ 50 balancing will be consider right whereas it is impossible or 60/40 should be right or other?
The degree to which class imbalance is a problem depends on both the properties of the dataset being examined and the prediction method being used. For example, if you are using SVM, then you can imagine that class imbalance may not be as big a problem, because classification is dependent on relatively few observations.
In some cases, class balancing is either not beneficial, or actually detrimental. For example, if you are using logistic regression, then balancing does not make sense. Although the loss used in optimization will be based on more of one class than another, it doesn't follow that this is detrimental to the estimates. It just means you'll have a better model of students that pass (in your case). In a two class problem, this is fine, because all the other students are the ones who fail. This has been shown empirically as well:
http://www.sciencedirect.com/science/article/pii/S0169207011001403.
There is also a good discussion here:
http://www.win-vector.com/blog/2015/02/does-balancing-classes-improve-classifier-performance/.
There are methods for which class balance can be a problem though, such as tree-based methods. There has been work done on balanced random forest, and it is also possible to use observation weights to achieve a similar result:
http://statistics.berkeley.edu/sites/default/files/tech-reports/666.pdf.
I think class distribution is not an issue may vary depending on nature of classifier. Once keep in mind, when you prepare development and validation set try to keep class distribution uniformity. So do some stratification before modeling.
- Badges
- Science topic