LightGBM Prediction Algorithm Based on Bayesian Optimization and Its Application

Under the background of the era of artificial intelligence, predicting product demand through machine learning can effectively improve the production and sales benefits of enterprises. Based on the shipment data of a large domestic manufacturing enterprise for dealers, this paper predicts the order quantity of products. Firstly, 597465 data from 1753 products are preprocessed and feature extracted, and 25 features including is _ holiday, Season _ 0, price _ bins, item _ code _ mean _ encoded, sales _chan_name are determined. The LightGBM algorithm is established to predict the product demand in the next three months, and the parameters of min _ child _ weight, lambda _ 12, feature _ fraction, bagging _ fraction, learning _ rate, seed, num _ iterations in LightGBM are optimized by Bayesian algorithm. According to the monthly grain size prediction, R2 is 0.92; Finally, the algorithm is compared. The prediction of BP neural network shows that R2 is 0.70. The experimental results show that the optimized LightGBM prediction result is significantly better than the BP neural network. The algorithm provides a scientific reference for the prediction of product order quantity.