neural network: probability of prediction

I have been using the neural network toolbox to predict the next value in a time series. This works, however I would like to know what is the probability of Matlab's neural network prediction.

E.g.: let's say we have time series [1 2 3 4 5]. The neural network would maybe predict the next value in the chain will be 5.9 (as an example).

Is there any (easy) way to derive the probability of this prediction? I would like to be able to tell something like this: it is predicted that the expected value of the next step will be 5.9 with a probability of 78%. It would be even better if I could get the entire probability distribution of the next value, or at least also the standard deviation. I hope anyone can help.

My point is that having a prediction without a probability does not help a lot, there might be a lot of uncertainty about this prediction. If one would forecast it is going to rain tomorrow this would give little information. If one would say it is going to rain with 95% probability than I know how to handle.

[Natanael silva]

I have the same problem, please if you had the solve, answer here, thank very mach

[Juan Klopper]

The .prediction method of your trained neural network takes the feature set (matrix) as input. The output is a prediction vector and a matrix of probabilities. Example below.

[YPred, YProb] = trainedModel.predictFcn(sim)

Answer 1

[Greg Heath]

The only way I know of to obtain error bars for a neural net output is to

 1. Assume an input probability distribution 

 2. Assume an input variance

 3. Create many random realizations

 4. Calculate the resulting spread of the prediction

Hope this helps.

Greg

Answer 2

[Morgan facchin]

Hello,

I am working on a similar problem and I am having difficulties, could you tell me how you would do to find that 5.9 in that particular example ??

(Sorry that is not an answer but your first step is what I am trying to do!)

Thank you !