Efficiency of super learner combining boosted regression tree, deep neural network, and frequency ratio models for mineral water potential mapping

Abstract:

Many learners such as boosted regression tree (BRT), deep neural network (DNN), and frequency ratio (FR) models are widely used as data-driven model by learning the observations. Their efficiency can be affected by the data distribution and separation to train-validation data, the optimal learner is determined differently depending on feature of research and observed data. ‘Super learner’ is a predictive algorithm combining candidate learners, and provides the highest level of performance and stability in all utilized learners. In this study, super learner was applied in groundwater potential mapping for mineral water, and its performance and stability were evaluated to be compared with those of the other candidate learners (BRT, DNN, and FR models in this research). Cross validated risks of learners were computed, super learner represented the highest performance and low variation of performance depending data for learning. Stability of learners was also confirmed in their estimation, super learner produced more stable estimated results compared with the others. Superiority of super learner to the candidate learners will lead to efficient data-driven modeling and high quality of estimation and prediction.

Acknowledgement: This research was supported by the National Research Council of Science and Technology(NST) grant funded by the Korea government(MSIP) (No. CAP-17-05-KIGAM)