How to predict the output from a classifier model?

I have used the classification learner to train image data using a SVM. The output shows it to be 95% and the confusion matrix looks as follows:

When I export the model and try to predict the Image features in matlab I get a confusion matrix as the one below, which seems weird and makes me believe I am doing something incorrect. Could someone please help?

% Training Data
imset = imageSet('HOG\Train_Diff','recursive');
bag = bagOfFeatures(imset);
imageFeatures = encode(bag, imset);

signData = array2table(imageFeatures);
signData.signType = getImageLabels(imset);

% Test Data
imsetTest = imageSet('HOG\Test_Diff','recursive');
bagTest = bagOfFeatures(imsetTest);
imageFeaturesTest = encode(bagTest, imsetTest);

signDataTest = array2table(imageFeaturesTest);
signDataTest.signType = getImageLabels(imsetTest);

%% Classification 
classificationLearner

%%  Test Data Confusion matrix
model = trainedModel.ClassificationSVM;
predictedLabels = predict(model, imageFeaturesTest);
testLabels = signDataTest.signType;
plotconfusion(testLabels,predictedLabels)