Sales Prediction Optimization via Gradient Boosting Decision Tree

Abstract

Sales prediction is always a top priority for brands, effectively reducing unnecessary waste under the essential circumstance that all customers’ needs have been met. With an efficient prediction model, brands can make accurate inventory decisions and get a significant increase in profit. This paper provides an efficient sales prediction model for cross-border e-commerce companies with Gradient Boosting Decision Tree. We use the sales volume in the past months, their statistical characteristics, difference characteristics and month on month characteristics as the characteristics we choose. We use the Mean Absolute Percentage Error perimeter to compare the accuracy of our method with the Moving Average Method to assess the performance of our model. The results show that our iterative model has higher accuracy compared to the Moving Average Method. It also has a significant edge in both automation and analyzing big-scale data sets, which means that it solves the low feasibility of existing methods. On this basis, we finally conclude that our machine learning model can make great contributions to sales prediction.