Customer-Induced Shipment Postponement in Make-to-Order Manufacturing: An ERP-Based Predictive and Prescriptive Production Planning Framework
Applied Sciences (Switzerland), cilt.16, sa.16, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 16 Sayı: 16
- Basım Tarihi: 2026
- Doi Numarası: 10.3390/app16168077
- Dergi Adı: Applied Sciences (Switzerland)
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Applied Science & Technology Source, Compendex, INSPEC, Directory of Open Access Journals
- Anahtar Kelimeler: CRISP-DM, Customer-Induced Shipment Postponement (CISP), delivery delay prediction, Explainable Artificial Intelligence (XAI), linear programming (LP), machine learning, make-to-order (MTO) manufacturing, predictive–prescriptive analytics, production planning, SHAP, Supply Chain Risk Management (SCRM)
- Yıldız Teknik Üniversitesi Adresli: Evet
Özet
Featured Application: A decision-support framework that uses ERP order data available at the time of order acceptance to predict Customer-Induced Shipment Postponement risk and support production planning before production release in make-to-order manufacturing. In make-to-order (MTO) manufacturing, customer-side issues involving payment, documentation, logistics, or site readiness may postpone shipment after order acceptance, increasing finished-goods inventory, capital tie-up, and capacity inefficiency. Although prior studies mainly address in-production delays or pre-order demand uncertainty, Customer-Induced Shipment Postponement (CISP) and its integration into production planning remain comparatively underexamined. This study proposes a predictive–prescriptive decision-support framework that uses ERP data from a real MTO manufacturer to predict shipment-timing outcomes associated with CISP and transfers the resulting predictions into an offline linear-programming-based production-planning model balancing capacity and inventory. Following CRISP-DM, 18,506 order records from 2017 to 2020 were analyzed, and eight machine-learning methods were evaluated across four scenarios combining Full and Parsimonious feature sets with three-class and binary target structures. Models were evaluated using an ex-post cost framework, with conventional classification metrics retained as diagnostic indicators. The selected model was interpreted using SHapley Additive exPlanations (SHAP). Under the stated evaluation assumptions, integrating predictions into the offline LP-based production plan reduced total planning cost by 22.85% relative to the no-prediction baseline. This suggests that the proposed framework can generate measurable planning value.