An innovative multi-factor prediction algorithm construction and importance analysis model based on GBDT-LightGBM

With the rapid development of machine learning and artificial intelligence, the research of prediction models is becoming more and more popular. This paper aims to propose a general prediction model and analyze the importance of features. First, this paper builds the GBDT(Gradient boosting decision tree) and finally, the LightGBM (Light Gradient Boosting Machine) algorithm based on the Bagging Ensemble algorithm. MAPE and R2 were set as test indicators, and 31 groups of data from 1992 to 2022 were selected for training and learning. The results show that: (1) The GBDT-LightGBM model constructed in this paper has good robustness, with MAPE of 9.658 and R2 up to 0.899. (2) GBDT-LightGBM prediction model can better evaluate the influence of independent variables on dependent variables. This study provides an effective and general methodology for forecasting models.