It doesn’t really matter which one you choose since this decision will not change the final results. It’s up to you to decide your favorite option to make predictions because the three options are equally valid.
You may want to predict question by question when you do not have all the information about which fields are more important so the model guides you step by step with the right next question to ask. This method is a good fit for those users that may not have familiarity with the underlying model. For instance, when a medical professional is trying to predict whether a patient is likely to have a certain disease the answers to the question will determine what is the next best question to ask until you reach the final prediction. Please note this type of prediction does not support text or items fields.
The predict option is quicker since you have all fields available and you can see upfront which fields are more important for the prediction.
The third option, batch predictions, allows you to upload a new dataset in the same format as the training data, and use your model to make predictions on the new data. Then, this new file can be scored against the model and the results can be downloaded easily. Batch prediction method is the quickest of all three as it lets you process thousands to even millions of new data instances in one operation.