Tensorflow classification with extremely unbalanced dataset

Question

I am using TensorFlow LinearClassifier and also DNN to classify two - classes dataset.

However, the problem is the dataset contains 96% of Positive output, and 4% of negative output, and my program always return the prediction as Positive. Of course, in this case I will achieved the accuracy of 96%, but it does not make sense at all.

What is the good way to deal with this kind of situation?

Answer 1

You could try changing the cost function so that a false positive output would be penalized more heavily than a false negative.