Sales Forecasting Based on LightGBM

The combination of data science and machine learning is making sales forecasting possible. This will help improve the competitiveness of retail companies. This paper is based on the LightGBM framework, which is an improved GBDT model to realize Wal-Mart sales fore-casting. Large amounts of data require preprocessing so feature engineering is performed in this paper. First remove some features that are not related to the model input. Then the features are extracted and classified, and the mean, standard deviation and other statistics of some features are obtained. Experiments results show that our method has an RMSE of 0.641, which is significantly better than Logistic Regression (0.803) and SVM (0.732). In addition, this paper also shows the 20 top feature importance. This is of great significance for guiding the company's sales