why do I have different results when i reuse the generated code from the classification learner app ?

Hello, I need some guide on how to reuse the generated code because it seems that i am making a mistake. I have used an input of 95x89 that i named it "trainerfeat" , where the 89's column is the response column,i fed this input to the app and chose 8% heldout validation , then i trained the data with svm and it turned out the the quadratic svm is the best and gives me 100% accuracy, then i chose generate the code option from the export model list and also chose the export model option. later, i have saved the generated code and saved the model as trainedclassifier4. then in the command line i wrote:

>> [trainedClassifier4, validationAccuracy] = trainClassifier(trainerfeat)
but it gives me  different accuracy results !:

>> [trainedClassifier4, validationAccuracy] = trainClassifier(trainerfeat)

trainedClassifier4 = 

             predictFcn: @(x)svmPredictFcn(predictorExtractionFcn(x))
      ClassificationSVM: [1x1 ClassificationECOC]
                  About: 'This struct is a trained classifier exported from Classification Learner R2016a.'
           HowToPredict: 'To make predictions on a new predictor column matrix, X, use: …'

validationAccuracy =

      0.5714

so can somebody tell me where i have gone wrong? any suggestions please?