The Weighted Multiple Meta-Models Stacking Method for Regression Problem
Dong Wang, Xishun Yue · 2019
When solving classification and regression problems using various machine learning models, we often choose the model that has the best performance. However, each model has advantages and disadvantages when it comes to the specific problem and scenario. If we combine the benefits of multiple models, then we will get better results. In the past, many methods for improving the prediction performance of various models are proposed. In this paper, we introduce a method called stacking. The stacking method introduces the concepts of base model and meta-model, which takes the predictions of base models as the input of the meta-model and the output of the meta-model as a final prediction. We have made some improvements to the stacking method. Instead of using single meta-model, we use the output of each base model as the input to multiple meta-models, then combine the outputs of multiple meta-models as a final output. The experimental results show that the weighted multiple meta-models stacking method outperform the traditional stacking method by reducing RMSE to 0.1245, while the best score of other models is 0.1292.