Class Weighting - Yousef's Notes

SVM, Decision Tree, Random Forest
For each class, we provide a weight and multiply the cost function by that weight.
e.g. use a greater weight for a minority class to keep it relevant during training.

Problem: distinguish between genuine and fraudulent e-commerce transactions.
SVM moves the hyperplane to reduce the number of misclassified examples
The examples of genuine transactions are much more frequent.
If the misclassification cost is the same for both classes, the “fraudulent” examples, in the minority, risk being misclassified to allow classifying more of the majority class correctly.
Solution: we use SVM with soft margin, defining a cost for misclassified examples.