Keras ANN Trained Model's Accuracy change on prediction

Question

I have trained an ANN Binary classifier using Keras. It gives 90% accuracy. After testing when I predict same data again but pass only one class then accuracy decreases to 40%. I have figured out that if I pass mixed classes while predicting then it will give me around 90% accuracy and if I pass data points of only one class then accuracy decreases .As I increase the data points of other class as well then accuracy increases. Long in short. CASE 1: 100 samples from class 0, 100 sample from class 1, on predicting using trained model Accuracy = 90%

CASE 2: Same 100 samples from class 0 passes to same trained Models give me 40% accuracy. Why accuracy changes ?

EDIT: I'm performing Standardization before predicting every time which effects the predictions. How to handle this case? Any suggestions would be much appreciated. Thanks

Answer 1

This problem is solved if we use same Mean and Standard Deviation which used for standardization of training samples.