While Normalizing/Scaling the training (or test) dataset, please note that the ‘label’ column (whose value needs to be predicted) should be dropped from the original training (or test) dataset. Hence, for ‘California housing prices’ dataset, please drop ‘median_house_value’ column from your training/test dataset, and then apply Normalization/Scaling on it. The predicted value that you get from ‘model.predict(scaled_training_dataset)’ will not be in normalized/scaled format, it will be in original format (that existed before Normalization/Scaling), and you can easily compare this predicted value with the ‘label’ value.
Hence, you will not have any need to convert the Normalized/Scaled data to original form. Anyways, the original dataset is also with you, because normalization/scaling will create a new ‘scaled’ dataset, original dataset will remain as it is.