Why the accuracy in testing phase of ANN difference for matlab 2010 and 2018

%Read the data file

pt=xlsread('Data normalize.xls');

p1=pt(1:893,1:4);

t1=pt(1:893,5);

%Make the transpose of the input and output pattern

p=p1';

t=t1';

rand('state',1);

%Set the network architecture

%Both the hidden and the output layer neurons are activated by binary sigmoidal activation function.

%net=newff(minmax(p),[Hidden Node Number Column Target],{'tansig' 'tansig'},'trainlm');

%Training parameters=Levenberg-Marquardt

net=newff(minmax(p),[10 1],{'tansig' 'tansig'},'trainlm');

%Set the training parameters

net.trainParam.epochs=j;

net.trainParam.goal=0.0100;

net.trainParam.lr=0.001;

net.trainParam.min_grad=1e-10;

net.trainParam.mu=0.9;

net.trainParam.mu_max=1.0e10;

%Train the network

net=train(net,p,t);

%Save the final weight and the network architecture in net

save net net;

% Validation

%Read the data file

p1=pt(1:893,1:4);

t1=pt(1:893,5);

[m1 n1]=size(p1);

p=p1';

t=t1';

%The data to simulate the network must be of column format

y=sim(net,p);

y';

round(y');

%Compare network response against the desired target response

[y' t' (y'-t')];

[round(y') t' round(y'-t')];

%Build the classification matrix which will provides a comprehensive picture of a classifiers performance

diff_1=[round(y'-t')];

A1=diff_1(:,1);

a_a=length(find((A1==0)));

fprintf('Number of epochs = %d\n',net.trainParam.epochs);

%Classification matrix in number

cm=[a_a m1];

fprintf('Number of Correct Classification : %f\n', (cm(1,1)));

fprintf('Number of Incorrect Classification : %f\n', (cm(1,2)-cm(1,1)));

%Classification matrix in percentages

cm_p=(cm ./ m1) .* 100;

fprintf('Percentage of Correct Classification : %f%%\n', 100*(cm(1,1))/m1);

fprintf('Percentage of Incorrect Classification : %f%%\n', 100*(cm(1,2)-cm(1,1))/m1);

%% Harumanis Classification

%Bayesian Regulation

%Not-healthy=0, Healthy=1

%clc;

%clear all;

%close all;

%% Testing the Classifier

%Read the master data file

pt=xlsread('Data normalize.xls');

[m n]=size(pt);

%Read the data file

p1=pt(894:1488,1:4);

t1=pt(894:1488,5);

[m1 n1]=size(p1);

p=p1';

t=t1';

%Load the final weight and simulate the output(y)

load net net;

%The data to simulate the network must be of column format

y=sim(net,p);

y';

round(y');

%Compare network response against the desired target response

[y' t' (y'-t')];

[round(y') t' round(y'-t')];

%Build the classification matrix which will provides a comprehensive picture of a classifiers performance

diff_1=[round(y'-t')];

A1=diff_1(:,1);

a_a=length(find((A1==0)));

%Classification matrix in number

cm=[a_a m1];

fprintf('Number of Correct Classification : %f\n', (cm(1,1)));

fprintf('Number of Incorrect Classification : %f\n', (cm(1,2)-cm(1,1)));

%Classification matrix in percentages

cm_p=(cm ./ m1) .* 100;

fprintf('Percentage of Correct Classification : %f%%\n', 100*(cm(1,1))/m1);

fprintf('Percentage of Incorrect Classification : %f%%\n', 100*(cm(1,2)-cm(1,1))/m1);

end

[boonchin ong]

This is my data file