adding noise increased accuracy after training

Hello 

I calculated noise in two different ways and each time i training im getting accuracy 100% for noise data and 99 for data without noise

can somone please advice me where i do mistake

train and target mat files avaliable under below link

https://drive.google.com/open?id=1CzfnIY5DZqcu8Vt-zxM3I6sVLnzLhwK5

purpose of noise is i would like to decrease data accuracy and then train on different parameters to compare training results

code

adding noise(50%)

TrainSetArray = load('TrainSetArray.mat')
TargetSet = load('TargetSet.mat')
%Add noise
noiseSignal = cos(5 * pi * 100 * TrainSetArray)+sqrt(5) * randn(size(TrainSetArray));
noiseSignal= num2cell(noiseSignal)
noiseSignal=cell2mat(noiseSignal)
TargetSet = Tbt(:,(563:563));%data.simplefitTargets';
TargetSet= table2array(TargetSet)

training

x = noiseSignal';
%x = TrainSetArray';
t = TargetSet';


trainFcn = 'trainlm';  % Levenberg-Marquardt backpropagation.

% Create a Fitting Network
hiddenLayerSize = 1;
net = fitnet(hiddenLayerSize,trainFcn);

% Setup Division of Data for Training, Validation, Testing
net.divideParam.trainRatio = 70/100;
net.divideParam.valRatio = 15/100;
net.divideParam.testRatio = 15/100;

% Train the Network
[net,tr] = train(net,x,t);

% Test the Network
y = net(x);
e = gsubtract(t,y);
performance = perform(net,t,y)

%checkCheck accuracy- noise set get 100%, without noise (train target) get 99%
[c,cm,ind,per] = confusion(t,y)
fprintf('Percentage Correct Classification   : %f%%\n', 100*(1-c));
fprintf('Percentage Incorrect Classification : %f%%\n', 100*c);

Answer 1

[Mahesh Taparia]

Hi

Noise addition acts as regularizer and it reduce overfitting. The model becomes more robust, so the accuracy of the model increases.

[Tomasz Kaczmarski]

thank you for explanation

in this case how i can modify raw data to decrase accuracy ?