Integrating a Novel Fuzzy Network DEA Method with Machine Learning Algorithms to Predict the Performance of Production Systems

Policymakers often face challenges when assessing the performance of multi-process production systems due to their complex structure and limited data availability. Furthermore, the interrelations among sub-processes and data processing, especially in imprecise data environments, introduce a variety of variables and constraints in models that evaluate system performance. State-of-the-art techniques, such as Machine Learning (ML) approaches, can be integrated with performance evaluation approaches to deal with the computational volume and enhance the accuracy of predictions. This paper, for the first time, presents a combination of a novel fuzzy network Data Envelopment Analysis (DEA) method with ML techniques to predict the performance of multi-process production systems with imprecise data. The fuzzy approach and its transformation to linear programming enhance the DEA method’s applicability and flexibility in handling imprecise and uncertain data. The proposed methods are applied to evaluate production periods in an Iranian canning factory, considered as distinct production systems. The results show that among three ML algorithms — Decision Tree (DT), Random Forest (RF), and K-Nearest Neighbors (KNN) — the RF algorithm has the highest accuracy in predicting efficiency scores, with a rate of 88%.