Construction and Application of Mathematical Model of Stacking Integrated Algorithm

Ensemble learning is a learning strategy that uses multiple independent learners to learn, and integrates their output results through some specific rules to obtain a strong learner. The integrated Stacking technology combines the predictions of multiple models and uses another machine learning model for final training. The primary learner used in this paper is AdaBoost, CART and KRR, and K-fold cross-check is used for training. The secondary learner is built on LightGBM model, which is more suitable for large-scale data. After the model is constructed, model parameters are optimized based on Sparrow Search to reduce the risk of overfitting or underfitting and find the best balance between model complexity and performance. After the model is constructed, data sets are selected for simulation experiments and three indicators MAE, MSE, and RMSE are used to evaluate the Stacking model. The results show that the performance of the stacking model is best compared with that of a single model.