Application of Stacking-Based Ensemble Learning Model for Water Quality Prediction

Abstract

Water is the source of life, and the growth of animals and plants cannot leave the water source. The quality of water will directly affect the life and health of humans, animals and plants. In order to predict the concentration and changing trend of various pollutants in water bodies and promote the comprehensive management of water resources, this paper proposes a new integrated model based on the idea of Stacking integrated learning. The model is based on XGBoost, support vector regression, and multi-layer perceptron. The model is constructed with ridge regression as the meta-model. The model was applied to the pH and total nitrogen content of water quality, and the mean absolute percentage error was used to quantitatively evaluate the prediction results, and the results of the ensemble learning model were compared with the prediction results of a single base model. The results show that the stacking ensemble learning idea can effectively improve the prediction ability and generalization performance of the base model. The proposed ensemble learning model has very good prediction ability and generalization ability, and has great potential in other prediction fields such as water quality.

Forecasting Albacore (Thunnus alalunga) Fishing Grounds in the South Pacific Based on Machine Learning Algorithms and Ensemble Learning Model

“…This study found that the ELM can integrate learning based on the output results of the base model as the input features of the second-layer meta-learner (LR) model, thus better integrating the advantages of the base model and producing more stable results. Meanwhile, different base models were 5-fold cross-validated, avoiding the problem of the poor generalization ability of test data division, effectively suppressing model over-fitting and improving the generalization ability and forecast accuracy of the ELM, similar to the findings of Cui et [20,50,51]. The XGBoost model, as a single machine learning algorithm, was better in forecasting albacore fishing grounds (ACC = 82.53%), where its overall accuracy was 1.58~15.09% better than that of other machine learning algorithms, and its accuracy (R 2 ) in forecasting high-yield tuna fishing grounds reached 90.37%.…”