I have a multi-class sentence classification problem and i want to achieve some sort of unknown class in prediction. There are n different labels and sometimes the sentence must not be classified or classified as unknown\None.
If you have some options to achive this behavior, i'd be very pleasured
A classifier can only predict what it sees in the training data. If you do not have an "unknown" in the training data, you can never see in at test time either. To make this problem even worse, neural networks tend to be very confident when making incorrect predictions.
If you have the instances belonging to unknown classes in your training data, then use it as any other class. If you don't, you might have a look at estimating classifier confidence.
