How to get r-square,mean absolute error and mean square error after train neural network?

Hi all, I train neural network as follow command

net.divideFcn = 'dividerand'
net.divideParam.trainRatio= 0.6; 
net.divideParam.testRatio= 0.2; 
net.divideParam.valRatio= 0.2;
[net,tr]=train(net,input,target);

I want to get r-square,mean absolute error and mean square error from train,test and validation data

Cloud you please advice ?

Answer 1

[Paras Gupta]

Hi Ninlawat,

I understand that you want to compute different network performance metrics on the train, test, and validation data after training a neural network object in MATLAB.

The following code illustrates one way to achieve the same:

% dummy data
input = rand(1, 100); % 1 feature, 100 samples
target = 2 * input + 1 + 0.1 * randn(1, 100); % Linear relation with some noise

% Define the feedforward network
net = feedforwardnet(10); % 10 hidden neurons

% Set up the data division
net.divideFcn = 'dividerand';
net.divideParam.trainRatio = 0.6;
net.divideParam.valRatio = 0.2;
net.divideParam.testRatio = 0.2;

% Train the network
[net, tr] = train(net, input, target);

% Get the network outputs
outputs = net(input);

% Separate the outputs for training, validation, and testing
trainOutputs = outputs(tr.trainInd);
valOutputs = outputs(tr.valInd);
testOutputs = outputs(tr.testInd);

% Separate the targets for training, validation, and testing
trainTargets = target(tr.trainInd);
valTargets = target(tr.valInd);
testTargets = target(tr.testInd);

% Calculate and display R-square, MAE, and MSE for each dataset
datasets = {'train', 'val', 'test'};
outputsList = {trainOutputs, valOutputs, testOutputs};
targetsList = {trainTargets, valTargets, testTargets};

for i = 1:length(datasets)
    dataset = datasets{i};
    outputs = outputsList{i};
    targets = targetsList{i};
    
    % R-square
    SS_res = sum((targets - outputs).^2);
    SS_tot = sum((targets - mean(targets)).^2);
    R_square = 1 - SS_res / SS_tot;
    
    % Mean Absolute Error (MAE)
    MAE = mae(targets - outputs);
    
    % Mean Square Error (MSE)
    MSE = mse(net, targets, outputs);
    
    % Display the results
    fprintf('%s R-square: %.4f\n', dataset, R_square);
    fprintf('%s MAE: %.4f\n', dataset, MAE);
    fprintf('%s MSE: %.4f\n', dataset, MSE);
    fprintf('\n');
end

You can refer the following documentation links for more infromation on the properties and functions used in the code above:

• mse - https://www.mathworks.com/help/deeplearning/ref/mse.html
• mae - https://www.mathworks.com/help/deeplearning/ref/mae.html
• divderand - https://www.mathworks.com/help/deeplearning/ref/dividerand.html

Hope this helps.