How to achieve better regression using Neural network having 364*6 inputs and 364*1 output?

Hi, I have 364*6 inputs and 364*1 output and loading into neural network of 10 neurons and all other settings are given below:

clear all; 
  close all;
load('nmtraindata.mat'); 
in=input';
ref=output';

    net = feedforwardnet(10);
    % net.performFcn = 'mse';
    net.performFcn = 'mae';

    %set NN function and other parameters 2if desired
          net.layers{1}.transferFcn = 'tansig'; % transfer function is "x=y" %tansig logsig
    % net.layers{2}.transferFcn = 'tansig';
          net.trainParam.max_fail = 1000; 
          net.trainParam.min_grad=1e-9; 
          net.trainParam.show=10; %showing intervals
          net.trainParam.lr=0.1; %learning rate
          net.trainParam.epochs=1000; %maximum iterations 
          net.trainParam.goal=0; %error goal
          net.trainParam.mc = 0.9;

    % train the NN
    net = train(net,in,ref); 
    estim_out = sim(net,in); % simulate the network to calculate output
    plot(estim_out,'r'); hold on;
    plot(ref,'b')

If there is less noisy data then regression plot is 1 (desired) but when i program large data set like the above one, i don't regression above 0.5. Am I missing something, like, do I need to change settings of neural network?

Looking forward for a positive solution and guidelines to this problem.

[Greg Heath]

I hope size(in) is not what you claim in the title and text.

Greg

[Ahsen Noor Subhopoto]

I apologize for the mistake in the question. Input is 364*6 and the output is 364*1. However, the question and figure remains the same; how to improve regression plot of given data set. I have tried different settings but it(regression) is getting worst by changing the settings.

[Greg Heath]

1. You mean 6*364 & 1*364 !

2. Use scatter plots, NOT line plots!

Greg

[KSSV]

Attach your data.

[Ahsen Noor Subhopoto]

Hi Greg and KSSV. I have attached my inputs plot (fig 1) and input vs output plot (fig 2). Output is 1 and 0.

Greg. I also tried alternatives to improve my R in regression plot of neural network. Unfortunately, I have to train my data 5000+ times to improve R from 0.3 to 0.7. According to you, we need to train data at least 20 times.

1. I don't understand, why my data is behaving different for the neural network. I used default settings of feedforwardnetwork for training it 5000 times.

2. Why NN was limited to R = 0.7 not 1 (desired)

3. What do I need to do to improve my NN regression plot to R = 1. you said use scatter plots but GUI of matlab is bydefault line plot.

kind regards,