NaN in inputs for test datasets in Neural network (patternnet)

Hi Sir/madam,

I trained an ANN with four input and one output. its a Classification and my output has three classes, I used one-hot coding. I used patternnet to train/test my ANN.

Everything is alright. Now question is lets say I have a missing value for one of my inputs like input matrix = [10, 20, NaN, 30]. When I enter this to get result (prediction) from ANN, its generates NaN as result! But, I heard that ANN should generate a result, its like a human brain.

Answer 1

[Sahil Jain]

Hi Ahmad, a neural network is essentially a series of matrix multiplications but with a non-linear activation after each matrix operation. As a result, if an input contains a NaN value, then it would be multiplied and added to other values during the operations and would result in a NaN output. You can read more about how a neural network works here.