Predicting Invoice Dilution in Supply Chain Finance with Leakage Free Two Stage XGBoost, KAN (Kolmogorov Arnold Networks), and Ensemble Models
New ML model uses Kolmogorov Arnold Networks (KANs) to forecast payment gaps in real-time supply chain finance.
Researchers Pavel Koptev, Vishnu Kumar, and team developed a novel AI framework combining leakage-free two-stage XGBoost, Kolmogorov Arnold Networks (KANs), and ensemble models to predict invoice dilution—the gap between approved and collected invoice amounts. The system analyzes nine transaction fields from production data to supplement deterministic algorithms. This enables real-time, dynamic credit limit projections for buyer-supplier pairs, moving beyond traditional irrevocable payment undertakings that hinder adoption.
Why It Matters
Reduces non-credit risk and margin loss in supply chains, enabling more flexible financing for sub-investment grade buyers.