<![CDATA[Invoice processing is a crucial component of a financial information system, particularly the accounts payable module integrated with an Enterprise Resource Planning (ERP) system. Delays in invoice processing are common, leading to financial penalties. This study employs a data-driven approach by integrating process mining and machine learning using the XGBoost algorithm to analyze the factors contributing to delays in the purchase-to-pay (P2P) workflow. Event log data from the BPI Challenge 2019 public dataset and refined to 3,471 complete cases. Process discovery and performance analysis were performed using the Disco tool, while an XGBoost binary classification model was developed with 16 process-level features extracted from the event logs. The model yielded an AUC-ROC of 0.6267 with a 5-fold cross-validation of 0.6382 and a standard deviation of 0.0099, indicating moderate discriminatory ability. Feature importance analysis identified process complexity (0.191), ideal workflow conformity (0.187), and invoice cancellation frequency (0.122) as the dominant delay factors, while document completeness contributed insignificantly. These findings suggest that in a complete P2P workflow, the cause of delay lies in the characteristics of process execution. The results of this study provide direction for developing accounts payable information systems with a focus on complexity reduction and real-time compliance monitoring.]]>
Copyrights © 2026
