Construction of Consumption Data Optimization Model Based on XGBoost Algorithm

In view of the difficulty in capturing the interactive effects and nonlinear relationships of factors affecting consumer purchase decisions and the imbalance of categories of purchase decision results, this article combines DNN (Deep Neural Network) and XGBoost (Extreme Gradient Boosting) algorithms to study the improvement of consumer purchase decision precision in financial and commercial big data, aiming to better understand consumer needs and improve marketing accuracy. First, transaction records and browsing history are obtained from an e-commerce platform and preprocessed. Then, a DNN multi-layer network structure is used to learn complex nonlinear relationships through nonlinear activation functions. In order to increase the prediction accuracy under unbalanced data, an ensemble of several decision trees is constructed and the parameters and loss functions are changed in conjunction with XGBoost. According to the experimental data, this model’s accuracy is 15.8%, 10.5%, 13.0%, 14.4%, and 8.2% higher than the control model’s, while its precision is 17.5%, 10.2%, 13.0%, 15.9%, and 10.3% higher. The conclusion shows that the application of DNN and XGBoost algorithms can help improve the understanding of consumer behavior and decision, and can provide new perspectives and basis for corporate marketing.