A Machine Learning Framework for De-Risking ERP Changes and Ensuring Financial Reporting Stability

ERP systems underpin financial operations in contemporary organizations, and even minor system changes can seriously compromise the accuracy of financial reporting. Conventional approaches to managing changes are highly reliant on subjective judgment and tend to not provide quantification of risk. This paper introduces a machine learning framework for predicting the impact that ERP change requests may have on financial reporting by using an 8,050-record dataset compiled across multiple implementations. Fifteen technical, procedural, and temporal features were engineered, standardized, and tested across four models: Support Vector Machine (SVM), Random Forest, Gradient Boosting, and Logistic Regression. Using stratified 10-fold cross-validation, the SVM outperformed other contenders with 99.19% accuracy, 98.9 % precision, and 99.3 % recall. Feature importance analysis identified Priority Level and Data Model Changes as the strong leading predictors of stability outcomes. The proposed framework establishes a quantitative and reproducible means of assessing the risk of ERP changes through a reduction in the level of system instability and improvement in the reliability of financial reporting. This work allows one to set a benchmark for data-driven ERP governance and underlines prospects for the real-time prediction of impact, platform-specific modeling, and integration into CI/CD-based workflows of change management.