I am working on a time series prediction problem. I am using keras models for machine learning.

For this prediction, weather variables are used as input. They can be of two types: forecasted and actual. I have acquired both these types of data of a considerable amount of time and I want to train and test my model on the data. My question is:

1) Should I use forecasted weather variables OR actual weather variables in input while training? (consider that only forecasted weather variables will be available at model inferencing time).

2) Same question 1 for testing.

Is there a rule or general practice regarding the above questions? If yes, I would like to know that.

  • $\begingroup$ What do you want to predict? Whether a given forecast will be right given past forecasts and past actuals? $\endgroup$
    – ignatius
    Apr 3 '19 at 9:30

According to me, for model training we should always use the actual data, so that your prediction is always close to realty.

But in case if I have the prediction data as well,

  1. I will trained another model with the prediction data.
  2. Analysis the outcome of the two models.
  3. Identify the deviation between two outcomes.
  4. Will use as threshold(variation) in future prediction.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.