Input and output in neural net toolbox

I want to show the effect of pH, Turbidity, Hardness, COD(Chemical oxygen demand), DO(Dissolved oxygen) on BOD(Biological oxygen demand) using the ANN toolbox. I have 18 sets of sample from different location. Will I take 6 inputs(pH, Turbidity, Hardness, COD, DO, BOD (6*18 matrix)) and One output (Desired values (6*1 matrix)) for constructing the work? Or will I take only the desired value of BOD as output? And how will I plot residuals vs. computed values of the parameters? I have attached our collected data.

Answer 1

[Greg Heath]

Please post Data.m or Data.txt

 [ 5 18 ]= size(input)
 [ 1 18 ] = size(target)
 MSE00    = var(target,1) % Reference MSE

% Documentation for designing the net

 help fitnet 
 doc fitnet

% However, the data set is so small, you need to minimize, by trial and error, the number of hidden nodes that yields a successful design. Typically, I use the goal

 MSEgoal = 0.01*MSE0
 net.trainParam.goal = MSEgoal

Since, by default, the initial weights and trn/val/tst data division are both random, a double loop design over numH hidden nodes

 j = 0
 for h = Hmin:dH:Hmax % numH = length(Hmin:dH:Hmax)
 j=j+1

and Ntrials number of random weights and random data divisions combinations

 for i=1:Ntrials  % Typically Ntrials = 10 or 15

will yield several successful designs. For examples, search

 greg fitnet

with Hmax, and/or Ntrials and/or MSEgoal.

Hope this helps.

Greg

[Yeasir Mohammad Akib]

Hey, Greg here is the txt file.