Συγκριτική αξιολόγηση τεχνητών νευρωνικών δικτύων και μηχανών διανυσμάτων υποστήριξης για την προσέγγιση της  αποδοτικότητας της περιβάλλουσας ανάλυσης δεδομένων

Γλεντής - Περιστέρης, Αριστοτέλης - Ηλίας

A comparative evaluation of artificial neural networks and support vector machines for approximating data envelopment analysis efficiency scores

Master Thesis

Author

Γλεντής - Περιστέρης, Αριστοτέλης - Ηλίας

Date

2026-09

Abstract

Data Envelopment Analysis (DEA) is the most widely used non-parametric method for evaluating the efficiency of homogeneous decision-making units (DMUs). This thesis proposes a novel DEA approximation framework based on machine learning techniques, specifically Artificial Neural Networks (ANNs) and Support Vector Machines (SVMs), with the objective of accelerating the efficiency evaluation process and reducing computational cost. More specifically, the proposed machine learning models are trained to estimate the efficiency scores generated by DEA. The datasets used for both training and testing are initially evaluated using DEA models, and the resulting efficiency scores serve as target values for the machine learning algorithms. A comprehensive set of experiments is conducted using synthetically generated datasets characterized by different data generation processes, varying numbers of inputs and outputs, and different numbers of decision-making units. Furthermore, the efficiency scores of the generated units follow specific statistical distributions, while the data are appropriately grouped into predefined intervals to ensure balanced representation. The experimental results demonstrate that machine learning models can approximate DEA efficiency scores with a high degree of accuracy. However, the findings indicate that Support Vector Machines consistently achieve superior and more stable performance than Artificial Neural Networks across all examined dataset sizes and distributions. In addition, the efficiency estimations produced by the machine learning models preserve the fundamental properties of the corresponding DEA models. Overall, the proposed approach can be effectively applied to large-scale datasets where traditional DEA methods become computationally expensive and impractical. At the same time, the study provides a foundation for future research in large-scale efficiency analysis, advanced neural network architectures, and broader artificial intelligence applications.

Postgraduate Studies Programme

Κυβερνοασφάλεια και Επιστήμη Δεδομένων

Department

Σχολή Τεχνολογιών Πληροφορικής και Επικοινωνιών. Τμήμα Πληροφορικής

Number of pages

Language

Greek

URI

https://dione.lib.unipi.gr/xmlui/handle/unipi/19774

Collections

Τμήμα Πληροφορικής

Show full item record

Except where otherwise noted, this item's license is described as
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα