Suppose you have n observations of a time series dataset. You split it up to n-k (train data) and k (test data) observations. You train a model using the train data and you can now make predictions. From what I understand multi-step forecasting means you want to predict the next q>1 observations. So you predict with your model the next q observations and you use a metric to compare them with their respective values from the test data. Now for the next step, will you need to train a new model which will also include the first q values from the test data and repeat the same process?
Asked
Active
Viewed 28 times
