The Rotate Stress of Steam Turbine Prediction Method Based on Stacking Ensemble Learning

A Fundamental Overview of SOTA-Ensemble Learning Methods for Deep Learning: A Systematic Literature Review

“…Therefore, it is relevant to know the individual combination strategies and their specifics to use them efficiently [28]. The ability to generalize includes a further goal of EL, namely the reduction of possible overfitting within the ensemble model [29]- [31]. Overfitting implies that during the training phase, the model learns the features and specifics of the training dataset extremely well but performs poorly within the test data or on data never seen before [10], [32].…”

“…Stacking or "Stacked Generalization" was proposed by Wolpert in 1992 [46]. Stacking is primarily like boosting, but unlike boosting and bagging, stacking is often used to combine different types of models [18], [31]. Stacking ensures the variation and generalization of the resulting ensemble through the different models, the basic idea being to identify training data that has not been learned correctly [18], [31].…”

“…Stacking is primarily like boosting, but unlike boosting and bagging, stacking is often used to combine different types of models [18], [31]. Stacking ensures the variation and generalization of the resulting ensemble through the different models, the basic idea being to identify training data that has not been learned correctly [18], [31]. During stacking, the entire training dataset is divided into M-blocks, with a distinction being made between training and test sets within the M-blocks [31].…”

“…Stacking ensures the variation and generalization of the resulting ensemble through the different models, the basic idea being to identify training data that has not been learned correctly [18], [31]. During stacking, the entire training dataset is divided into M-blocks, with a distinction being made between training and test sets within the M-blocks [31]. In the first instance, the entire training dataset is divided into training and test sets, and then the training datasets are further divided using "cross validation" [28].…”

“…In the first instance, the entire training dataset is divided into training and test sets, and then the training datasets are further divided using "cross validation" [28]. The cross-validation has the peculiarity of stacking that a combination takes place subsequently instead of the "winner-takes-it-all" approach [18], [31]. Cross-validation is used to increase computational efficiency and to avoid possible overfitting [31].…”