Abstract

Procurement inefficiencies and revenue leakage represent significant challenges in industrial supply chains, often resulting from contract non-compliance, price deviations, supplier irregularities, and operational inefficiencies. Traditional audit-based systems are largely reactive and insufficient for detecting complex, high-dimensional patterns embedded in modern procurement data. This study proposes a mathematically grounded and machine learning–driven framework for detecting procurement inefficiencies and minimizing revenue leakage. The approach integrates structured data from purchase orders, invoices, and contracts into a high-dimensional feature space and introduces a Procurement Inefficiency Index (PII) to quantify deviations across pricing, delivery, and transaction behavior. A hybrid modeling architecture combining supervised learning (XGBoost and Random Forest) and unsupervised anomaly detection (Isolation Forest) is developed to capture both known and emerging leakage patterns. The framework further incorporates an optimization model to minimize total leakage subject to contractual and operational constraints. Experimental results demonstrate that XGBoost achieves the highest performance with an accuracy of 0.94 and F1-score of 0.915, while anomaly detection enhances the identification of previously unseen inefficiencies. Sensitivity analysis reveals a trade-off between detection rate and false positives across varying thresholds, enabling adaptive decision-making. The findings confirm that the proposed framework significantly improves detection accuracy, supports real-time monitoring, and provides quantifiable metrics for procurement performance evaluation. This research contributes to the integration of machine learning and optimization in supply chain analytics and offers practical implications for enhancing transparency, cost control, and risk management in industrial procurement systems.