I designed a neural network classifier yielding 0 or 1. can i design with target value 1 alone.
1 view (last 30 days)
neural network was designed with 6 features each dataset. first 3 dataset have target value as 1. the last three datasets have target value 0. so if i give a sample after training, the output may be 1 or 0. now i do not have the features for last three sets. so that i may have only the first 3 sets , which has the target value 1. while running this tranined network, it always gives value 1 as output for all type of sets. is NN a binary classifier?
John D'Errico on 2 Jan 2017
Edited: John D'Errico on 2 Jan 2017
So, none of the data you have to train with has a target value of 0? How would you expect any modeling tool to recognize when the alternative applies?
If you have severely incomplete data, nothing will improve it, except for getting more useful data.
Think of it like this: a neural net (like any other modeling tool) needs to learn from data. If all the data you give it only EVER has the value 1, then it will learn that the prediction is ALWAYS 1. ALWAYS, for any set of inputs.