| Abstract [eng] |
The construction industry faces challenges due to urbanization, and sustainability standards. Furthermore, due to recent modernization and technological advancements, community needs have increased which directly influences construction operations. This has increased Supply chain delays, particularly in circular supply chains (CSC), impacting project schedules, financial targets, and sustainability programs. Challenges include supplier dependability, logistical efficiencies, policy and regulations, and market stability. However, current frameworks fail to address disruptions and therefore stress to incorporation of predictive analytics, highlighting the need for innovative approaches to increase CSC resilience and flexibility. The aim of the research is to design a hybrid framework that combines the Fuzzy Analytic Hierarchy Process (Fuzzy AHP) and explainable machine learning (XML) to handle important disruption causes in circular supply chains. The objective is to establish a single decision-making and forecasting tool that merges expert judgment with data-driven insights to manage disruptions and boost operational efficiency. Fuzzy AHP was utilized to identify and prioritize significant disruption causes, allocating the greatest weight to “logistics and operational efficiency”, followed by “supplier and material reliability”. Among sub-criteria, “financial stability” scored as the most crucial aspect, followed by “transportation reliability” and “warehouse efficiency and inventory management”. These findings were included in a machine learning model developed with XML algorithms which are gradient boosting (GB), decision trees (DT), and random forest (RF). The comparative analysis of the GB, DT, and RF models against the fuzzy AHP methodology gives a valuable understanding of their efficacy in anticipating supply chain disruptions. Each model's findings reveal distinct strengths and limits in terms of precision, recall, F1-score, and alignment with fuzzy AHP global weights, finally indicating the best acceptable approach to data-driven MCDM. The GB model emerged as the most robust, obtaining a high AUC of 0.91 and improved performance across accuracy, recall, and F1-score criteria. Its capacity to handle unbalanced datasets and complicated non-linear interactions allows it to recognize trends in both disruption and non-disruption instances efficiently. The DT model, although interpretable, displayed a modest AUC of 0.50 and failed to reliably predict disruption instances, as evidenced by its lower recall (33%) and F1-score (35%) for the “Disruption” class. The RF model fared marginally better than the DT, obtaining an AUC of 0.55, but its performance remained worse than GB, notably in recall for the “Disruption” class, demonstrating limited capabilities in recognizing disruptions reliably. This study is crucial since existing CSC frameworks typically fail to incorporate qualitative and quantitative methodologies, limiting their capacity to proactively detect and manage disturbances. By integrating Fuzzy AHP’s interpretability with XML’s predictive capacity, the hybrid framework overcomes this gap, delivering a solid, explainable strategy for decision-making. The results indicate that addressing supplier financial stability, streamlining logistics, and matching regulatory procedures are crucial for strengthening supply chain resilience and promoting circular economy objectives. In conclusion, this research adds to the literature by presenting a unique hybrid framework that combines Fuzzy AHP and XML to increase decision-making and prediction capacities in CSC. The findings illustrate the need to connect expert-driven goals with data-driven insights for sustainable supply chain management. Future studies must concentrate on adding real-time data, extending the framework to incorporate developing risk variables and verifying the technique across sectors and in different geographies. |
