Clustering with machine learning and using NDEA in development planning: A case study in the petrochemical two-stage SSC

This research utilizes Machine Learning (ML) to improve the evaluation of Decision-Making Units (DMUs) in a two-stage sustainable supply chain in the petrochemical industry in Iran. Efficiency calculations were conducted for ۲۸ units over ۹۰ time periods. The inputs and outputs of the supply chain were selected according to sustainability criteria, facilitating accurate estimations for production planning and unit development through a hybrid proposed method utilizing both ML and Network Data Envelopment Analysis (NDEA). The goal is to use ML clustering methods alongside network NDEA models to determine the most effective clustering algorithms for categorizing homogeneous units. The primary goal of our research is to leverage ML techniques to enhance the accuracy of decision-making processes, specifically in the clustering of similar units, to assess efficiency. The main goal is to create strategies for improving the performance of inefficient units by comparing them to the most efficient units in each cluster. By implementing the Deep Embedded Clustering (DEC) algorithm, we have discovered substantial improvements in efficiency assessment and development planning. The contrast between clustering outcomes and the traditional NDEA model highlights the importance of clustering in assessing the proximity to the efficient frontier and enabling focused development strategies. This study emphasizes the significance of employing ML for clustering to improve efficiency evaluations and strategic planning for sustainable development in industrial facilities. The results showed that the use of clustering for assessing the relative efficiency of units compared to the non-clustering method with DEA could averagely reduce ۱۸% of the distance of inefficient units from the cluster's efficiency frontier, representing a more attainable ideal goal for inefficient units.

کلیدواژه ها:

Clustering ، Machine Learning ، sustainable supply chain management ، Data Envelopment Analysis