Machine learning prediction and interpretability analysis of the association between digital transformation and enterprise performance: Based on the XGBoost-SHAP integrated framework

In the digital economy, the correlation between digital transformation and enterprise performance has gained growing attention. This relationship is complex, nonlinear and stage-dependent, which cannot be fully captured by traditional linear models. Existing studies seldom discuss the predictive contributions of different transformation dimensions and the interactions among technological foundation, digital business application and organizational change. This study selects Chinese A-share listed manufacturing firms from 2019 to 2023 and constructs an XGBoost-SHAP integrated framework for performance prediction and interpretation. We conduct standard data preprocessing and retain original transformation dimensions and control variables rather than adopting PCA. XGBoost is optimized with GridSearchCV, and SHAP is used for feature importance evaluation, local sample explanation and interaction analysis. The model achieves an R² of 0.85 on the test set, presenting satisfactory predictive performance. Results show digital business application makes the largest contribution, followed by technological foundation and organizational change. The two leading dimensions also generate strong interactive effects. Quantile analysis indicates transformation dimensions produce negative contributions at low transformation stages and remarkable positive contributions at high stages. These findings reflect predictive correlations instead of strict causal effects, and provide empirical support for resource arrangement and differentiated policy-making in manufacturing enterprises.