Research on Consumer Purchasing Prediction Based on XGBoost Algorithm

Abstract

To predict how many consumers will buy goods in the next month helps the e-commerce platform discover potential buyers and carry out the corresponding strategic activities. After analyzing and cleaning the data, we select user purchase features to use eXtreme Gradient Boosting (XGBoost) algorithm to train the divided data sets. Meanwhile, we choose Light Gradient Boosting Machine (LightGBM), Long Short-Term Memory (LSTM) and Fully Connected Neural Network (FCNN) as comparison algorithms. Expectedly, the experiments indicate that using the XGBoost algorithm to predict purchasing can improve performance. Specifically, LightGBM and LSTM increase significantly before remaining stable, whereas FCNN begins in the highest number falling dramatically to approximately the accuracy of 0.32 and keeps steady. Throughout the iteration process, the accuracy of XGBoost surpassed FCNN, and experienced a moderate increase from 0.55 to 0.67, increasing the accuracy by 12%.