Question

Consider the following setting. You are provided with n training examples: (x1, y1), (x2, y2), · · · , (xn, yn), where xi is the input example, and yi is the class label (+1 or -1). However, the training data is highly imbalanced (say 90% of the examples are negative and 10% of the examples are positive) and we care more about the accuracy of positive examples. How will you modify the perceptron algorithm to solve this learning problem? Please justify your answer.

Answer 1

ANSWER:

GIVEN THAT:

To consider the following setting. You are provided with n training examples: (x1, y1), (x2, y2), · · · , (xn, yn), where xi is the input.

There are many solution paths available for this problem. Some of them are :

i).Oversampling Minority Class:-
The minority class is used multiple times to train the perceptron. This way the minority input is amplified.

ii).Undersampling Majority Class:-
Remove some of the training tuples from majority class on random. This will reduce the excessive effect of majority class on the perceptron model.

iii).Modify Algorithm to be sensitive to Minority classes:-
Even a single misclassification in minority class has a huge cost because there are already very few cases of it.

Increase the misclassification costs