Your code and explanation is lacking, but if I have to guess Its probably because of the Normalize operation. When using Normazlise on the test-set, the parameters of the normalization are determined by the test-set only.

You should 'fit' a normalizer on the train-data and use this fitted normalizer to transform your test data.

Your code and explanation is lacking, but if I have to guess Its probably because of the Normalize operation. 
When using Normazlise on the test-set, the parameters of the normalization are determined by the test-set only.

You should 'fit' a normalizer on the train-set and use this fitted normalizer to transform your test-set.