Splitting of Dataset:
Dataset = Train1 + Test1
EEMD(Train1) = train1 + test1
I am forecasting on time series data("Dataset") using SVM. First I found the Intrinic Mode Function(IMF) of time series data (training dataset, Train1) using Ensemble Empirical Mode Decomposition (EEMD).Let's say I got 8 IMF (including resudial). Test dataset(Test1) is kept apart for only testing with final prediction.
Now I divided each IMFs into training(train1) and testing dataset(test1). Then trained 8 different SVM (SVR) model (8 IMFs and 8 SVR model) using sklearn (as it is shown in image - step 2)
Then forecasted as many data points as the number of elements in test dataset (Test1) using corresponding SVR model and IMF dataset (train1) on which it was trained. So now I have forecasted on each IMF. Step 2 is complete.
For step 3, I am confused.
Confusion 1: What dataset(either Train1 or train1) should I use for training the SVM in step 3? Or Should I use one of the 8 (already) trained model from step 2.
Confusion 2: It is written following for doing step 3 in the paper-
Use SVM to combine the prediction result of each IMF and residual component and generate an aggregated result as the final prediction of the original time series
How can I "combine" using SVM here? Does it mean to use forecasted data from step 2 as 8 input features (X) for SVM in step 3? What should be input(X) and target(y) for training SVM in step 3. I have never done ensembling before.
Following is the screenshots from the paper: