Product marketing prediction based on XGboost and LightGBM algorithm

Abstract

The XGboost and LightGBM algorithm performs predictive analysis of sales volume in the product sales data set. The principle of XGboost and LightGBM algorithm is studied, the predicted objects and conditions are fully analyzed, and the algorithm parameters and data set characteristics are compared. The results show that n_estimators have a small effect on the prediction of model XGboost, while gamma has a large effect on the prediction of model XGboost. Learning_rate has a small impact on LightGBM prediction, while n_estimators have a large impact on LightGBM prediction. Finally, the optimal parameters were obtained, and the sales volume from January to October 2015 was predicted based on the optimal parameters, and RMSE values of the two algorithms were obtained. Statistical analysis shows that there is no significant difference between the two algorithms in the optimal prediction results after adjusting their own parameters.