How to match predicted class and prabablity with actual labels

Question

I have trained a deep learning model based on Bidirectional LSTM and dense layer output. It is quite confusing that which output probability acquired using model.predi