As the paper says: "the instances extracted from the same size of win- dows will be used to train a complete-random tree forest and a random forest, and then the class vectors are generated and concatenated as transformed features."
Therefore the algorithm has to create the MGS estimators for each generated window.
As the paper says: "the instances extracted from the same size of win- dows will be used to train a complete-random tree forest and a random forest, and then the class vectors are generated and concatenated as transformed features."
Therefore the algorithm has to create the MGS estimators for each generated window.