Label Values for Training

[armlesscoder98]

Hi all! In the literature, I read many papers on the direct strategy; however, there is a gap in the information and one of the papers also identifies these gaps as follows: "In direct strategy, there is a lack of input information." Again, in the literature, papers generally mention the direct strategy as follows: it requires independent model for each prediction horizon and all the model are using the same historical data. However, there is no information about that historical data in the target values perspective in other words "labels". Let's assume that I will make a prediction of y(t+1), and y(t+2). So, am I going to use same samples for training my model or should I create other target values for my prediction model y(t+2)? Let's assume that I have an array like given in the figure [10, 20, 30, ... ,90] and I will look back 3 steps. So, to be able to make a prediction of y(t+1) and y(t+2) should I create train and targets as given in the figure or do I need to use approach (same samples for training my model) like in the Model 1 for each prediction horizon?

Thanks in advance!

[armlesscoder98]

If someone has the same confusion as I do, I found the answer: the strategy of the second approach, which is located on the right side of the figure, is the correct approach for the direct strategy!