normalization command before using classifier in matlab - matlab

Question

I am using normalization command:
normA = Xtrain - min(Xtrain(:));
Xtrain = normA ./ max(normA(:));
normB = Xtest - min(Xtest(:));
Xtest = normB ./ max(normB(:));
to normalized my data before using classifier (design tree), but every time I got very poor accuracy, it is around 55.00. Meanwhile, I got accuracy 93.88 without using the normalization algorithm. can anyone tell me what the problem exactly and what I have to do?
This is my code:
load('train_and_test_data.mat')
Xtrain= Xtrain(:, 2:42);
Xtest= Xtest(:,2:42);
normA = Xtrain - min(Xtrain(:));
Xtrain = normA ./ max(normA(:));
normB = Xtest - min(Xtest(:));
Xtest = normB ./ max(normB(:));
Mdl = fitctree(Xtrain ,Ytrain);
y =Mdl.predict(Xtest); %test
Conf_Mat = confusionmat(Ytest,y)
This small sample of data I am using before normalization:
1 0 0 0 0
17 4 2 2 0
38 20 17 0 0
11 2 2 0 0
2 1 1 0 0
11 1 4 0 0
8 5 1 1 1
21 1 16 0 0
27 12 11 0 0
13 11 2 1 0
12 3 2 2 1

Answer 1

You are not normalizing the training and the test set using the same transformation.
normA = Xtrain - min(Xtrain(:));
Xtrain = normA ./ max(normA(:));
normB = Xtest - min(Xtest(:));
Xtest = normB ./ max(normB(:));
Subtracting a different amount. Dividing by a different amount. Therefore the inputs from your test set are not comparable to the inputs from your training set. Instead, normalize your test data using the same transformation.
normB = Xtest - min(Xtrain(:));
Xtest = normB ./ max(normA(:));