KDD data set normalization

pls i need you guys help on how to normalize the KDD cup data set, or is some thing wrong with the code i have below.

% read excel file file from computer directory.

     INPUT_P= xlsread('C:\Users\5GPROGRAMER\Desktop\WORKING DIRECTORY\PROJECT WORKING DIRECTORY\TRAINING DATA\TRAIN DATA1.xlsx');
   TARGET_T= xlsread('C:\Users\5GPROGRAMER\Desktop\WORKING DIRECTORY\PROJECT WORKING DIRECTORY\TRAINING DATA\OUTPUT BINARY CODE2.xlsx');
   TEST_DATA= xlsread('C:\Users\5GPROGRAMER\Desktop\WORKING DIRECTORY\PROJECT WORKING DIRECTORY\TRAINING DATA\sample test.xlsx');

   % convert excel file to matrix ivp and ivt
   ivp=INPUT_P';
   ivt=TARGET_T';
   testinputs=TEST_DATA';

% start data set normalization

[pn,minp,maxp,tn,mint,maxt]=premnmx(ivp,ivt);

% creat neural network

 net=newff(minmax(pn),[5 5],{'tansig','purelin'},'trainrp')

[Stephen]

Am working on neural network using Back propagation to detect threat, and am working with the KDD data set. But each time i test the neural network with the test data set, its not giving the accurate result. Below is the full code of the work i have been doing.

% read excel file file from computer directory.
   INPUT_P= xlsread('C:\Users\5GPROGRAMER\Desktop\WORKING DIRECTORY\PROJECT WORKING DIRECTORY\TRAINING DATA\TRAIN DATA1.xlsx');
 TARGET_T= xlsread('C:\Users\5GPROGRAMER\Desktop\WORKING DIRECTORY\PROJECT WORKING DIRECTORY\TRAINING DATA\OUTPUT BINARY CODE2.xlsx');
 TEST_DATA= xlsread('C:\Users\5GPROGRAMER\Desktop\WORKING DIRECTORY\PROJECT WORKING DIRECTORY\TRAINING DATA\sample test.xlsx');

   % convert excel file to matrix ivp and ivt
   ivp=INPUT_P';
   ivt=TARGET_T';
   testinputs=TEST_DATA';
[pn,minp,maxp,tn,mint,maxt]=premnmx(ivp,ivt);
 net=newff(minmax(pn),[5 5],{'tansig','purelin'},'trainrp')

 % learning rate parameter for neural network
net.trainParam.epochs=100;
 net.trainParam.lr=5.0;
 net.trainParam.mc=0.6;
 %train neural network (net).
net = train(net,pn,tn);

out = sim(net,testinputs);

[dummy, I]=max(out);

if (I == 1 ) h = msgbox('ATTENTION: NORMAL PACKET','Description','none'); elseif (I == 2) h = msgbox('ATTENTION: NEPTUNE ATTACK','Description','none'); elseif (I== 3) h = msgbox('ATTENTION: REMOTE TO LOCAL ATTACK','Description','none'); elseif (I== 4) h = msgbox('ATTENTION: PROBE ATTACK','Description','none'); elseif (I== 5) h = msgbox('ATTENTION: USER TO REMOTE ATTACK','Description','none'); else h = msgbox('UNCLASSIFIED ATTACK','Description','none');

Answer 1

[Greg Heath]

1. I do not see where you have confirmed that you have obtained acceptable results from training. Where do you compare training output trainout = sim(net,ivp) with the training target ivt? Is

NMSE = mse(ivt-trainout)/mean(var(ivt',1)) << 1 ?

2. To use the testinput you have to normalize it, use sim, then unnormalize the answer.

Hope this helps.

Thank you for formally accepting my answer

Greg

[Greg Heath]

What version of the Toolbox do you have? That version of NEWFF has been obsolete for many years. The more recent version is also obsolete. The current MLP for regression and curvefitting is FITNET.

[Stephen]

the problem i have is the normalization, am not sure if this is correct [pn,minp,maxp,tn,mint,maxt]=premnmx(ivp,ivt);

[Stephen]

am using Matlab 7.8.0 (R2009a)

Answer 2

[Greg Heath]

Read the documentation on premnmx. Not only is it obsolete, it has a bug. Use mapminmax instead

help premnmx

doc premnmx

help mapminmax

doc mapminmax

Hope this helps.

Greg

PS. Always read the documentation if you have traced a problem to a particular function. If push comes to shove, you can always read the source code using the command type.

[Stephen]

thanks very much i used mapminmax() and it worked