Discrete Data Fitting using neural networks in matlab

Dear Friends,

I want to predict machine fault size (Continuous Variable) from 5 faults observation (Discrete Variables)

is it correct to use the fitting toolbox in MATLAB when my output data are Discrete??

when i use the fitting toolbox, the regression showing me R^2 = 0.92 (which is very good result) however, the MSE showing me very large number, the best MSE was =65 !! after few training.

shall i consider the MSE ?? if yes,, how to minimize it??

Regards

[Brendan Hamm]

You say what you are trying to predict is a continuous variable, so why would your output by discrete? It seems your predictors (inputs) are discrete based on what you said.

Anyhow, the size of the Mean squared error is really going to depend on the units in which a quantity is measured. That is if my response variable is measured in nanometers my MSE will be much larger than if measured in kilometers. Alone it means little, but can be useful as a comparison.

[salah mahdi]

<<

>>

thank you Sir,

I also dont know how to interprate this figure.

[priya singh]

how to feed discrete data in neural network ? Kindly help if you can provide any source to feed the continuous variable in discrete form.

Answer 1

[Greg Heath]

Normalize mse(error) for error = output-target with the mean target variance. Then

 vart = mean(var(target',1)) 

 NMSE = 1 - mse(error)/vart

and

 Rsquare = 1 - NMSE

Hope this helps.

Thank you for formally accepting my answer

Greg