Τεχνικές προεπεξεργασίας και εκτίμησης DEA αποδοτικότητας

Τσούνης, Αργύρης - Ρήγας

Preprocessing and approximate scoring DEA

Master Thesis

Author

Τσούνης, Αργύρης - Ρήγας

Date

2026-01

Advisor

Κορωνάκος, Γρηγόριος

Keywords

Data Envelopment Analysis ; Large-scale optimization ; MaxRatio method

Abstract

This thesis studies preprocessing and approximate scoring techniques to accelerate Data Envelopment Analysis (DEA) for large-scale datasets. DEA evaluates the efficiency of decision making units (DMUs) using linear programming, but its computational cost grows rapidly with dataset size, making direct application difficult in Big Data settings. The main contribution of the thesis is the development of MaxRatio method. It is based on an input–output ratio analysis to approximate DEA efficiency scores without solving LPs. To address the exponential growth of ratio combinations in the MaxRatio method and reduce the computational effort, sampling techniques are investigated to evaluate and confirm its effectiveness. The use of a limited number of ratios by the MaxRatio method through sampling led to important findings for understanding the relationships among ratios. In particular, it showed a strong correlation between ratio-based measures and DEA scores, and demonstrated that many ratio combinations lead to the identification of the same DMUs. Also, the MaxRatio method is compared to the preprocessing method KZCT-2019, which aims to quickly identify efficient DMUs. Experimental results show that the MaxRatio method outperforms the KZCT-2019 method in the number of efficient units it identifies as well as in the approximation of efficiencies, however it has higher complexity. Overall, both preprocessing methods identify a satisfactory number of efficient DMUs and provide reliable efficiency estimation. The results of this study are easily comparable with those of other studies and enable reproducibility and independent verification by other researchers.