Prediction and Characteristic Analysis of Enterprise Digital Transformation Integrating XGBoost and SHAP

Abstract

Objective: An interpretability model of enterprise digital transformation that integrates XGBoost and Shapley additive explanations (SHAP) is proposed to accurately identify the important factors that affect the digital transformation of enterprises and their mode of action, improve the digital capabilities and levels of enterprises, and prevent the risks of digital transformation of enterprises. Method: The annual report information of listed companies from 2009 to 2021 is used as the research object. First, the digital transformation index is constructed using the text mining method. Second, an enterprise digital transformation prediction model based on XGBoost is constructed and compared it with other mainstream algorithms, such as linear regression and random forest, to find a comprehensive optimal model. Finally, the SHAP interpretation framework is introduced to quantify and attribute the importance of each characteristic variable. Results: The results found that the XGBoost model outperformed the compared models in the mean absolute error and R2 performance indicators. In addition, development capability, comprehensive capability, and solvency are important characteristics influencing the digital transformation of enterprises, and they differ in the way, direction, and strength of influence on the digital transformation of enterprises. Research value: This paper applies XGBoost integrated learning method to identify the factors of enterprise digital transformation, which enables enterprises to assess their digital transformation status, discover the key determinants of digital transformation, and adopt effective digital transformation modes for higher value.