Combining machine learning techniques with NDEA methodology - the use of R.F. and A.N.N.

Aug 4, 2026

The work integrates Network Data Envelopment Analysis (NDEA) with machine learning and artificial neural networks to improve modelling under uncertainty. A machine learning step is applied upstream of NDEA by using Random Forest regression to adjust input/output data, sub-process resource allocation preferences, and cost, revenue, and profit targets influenced by internal and external factors. A downstream neural network step then optimizes the calculation of economic quantities derived from optimal NDEA solutions.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages17

Reading Time43 min

AccessFree to download

Abstract

Introduction

NDEA and network-configured production processes

Data requirements and uncertainty sources

Machine learning and neural networks integration

Munich Personal RePEc Archive Combining machine learning techniques with NDEA methodology: the use of R.F . and A.N.N . Pinto, Claudio University of Salerno 7 September 2025 Online at [https://mpra. ub. uni-muenchen. de/126539/](https://mpra. ub. uni-muenchen. de/126539/) [MPRA Paper No. 126539](MPRA Paper No. 126539) , [posted 27 Oct 2025 08:53 UTC](posted 27 Oct 2025 08:53 UTC) Combining machine learning techniques with NDEA methodology: the use of R.F. and A.N.N. Working Draft – Not for citation Claudio Pinto University of Salerno Abstract The objective of the present work is to integrate NDEA approach with machine learning techniques and neural networks. At this end we exploit the models proposed in Pinto, 2024. The integration process involves the application of a machine learning technique upstream of the resolution of NDEA models and the application of an artificial neural network downstream the resolution of a NDEA models. In particular here we propose the application of a Random Forest algorithm in regression models to adjust data on: 1) input and output, 2) resource allocation preferences among sub-processes, 3) cost budgets, revenue targets and profit targets, from the influence of internal and external factors in order to improve the calculation of optimal weights. Downstream of the resolution of NDEA models, the use of several artificial neural network models is to prosed to optimise the calculation of the economic quantities of interest derived from optimal NDEA solutions. The approach enhances the discrimination power and robustness of optimal NDEA weights as well as the robustness of the calculation of formulas ofthe economic quatities. Keywords: Network Data Envelopment Analisys, Random Forest Regression, Artificial Neural Network, external factors JEL code: C45,C53, C61,C67,L20 1. Introduction Network Data Envelopment Analysis (NDEA) (Cook, 2008; Kao, 2014) is a non-parametric methodology developed as an advancement of the well-known Data Envelopment Analysis (DEA)(Cooper, 2011) . Its characteristic is that it can take into account the internal structure of production processes that are presented as network systems. Using this methodology, a production process is modelled as consisting of parts interconnected by shared, intermediate and backward variables [(Castelli & Pesenti, 2014) (Castelli, Pesenti, & Ukovich, 2010) (Castelli, Pesenti, & Ukovich, 2001)] . Both methodologies require input and output data on the production technology of the production process for their application. Evidently, NDEA requires more information on, for example, how inputs and outputs are used by and between sub-processes, what the resource allocation preferences are between sub-processes, what the structural articulation of the production process is, how many sub-processes there are, whether or not there is feedback on final outputs, how many intermediate outputs there are, which inputs are shared between sub-processes, and so on (Castelli & Pesenti, 2014) (Castelli, Pesenti, & Ukovich, 2010) (Castelli, Pesenti, & Ukovich, 2001) . Consequently, solving an NDEA model also requires a greater amount of data. In reality, productive resources, i.e. the inputs of DEA or NDEA models, can be influenced by factors internal and external to the production process that make their availability and use uncertain or in any case variable. Just as in reality, the production of goods and services (the final outputs of DEA or NDEA models) can be influenced by internal and external factors that make both the quantity produced and their availability uncertain or in any case variable. And just as in reality, the level of intermediate output production may be subject to uncertainty or change. In the case of network-configured production processes, uncertainty or variability may also affect other model parameters, such as the allocation of resources among different sub-processes, the proportion of feedback

FAQs

What is the main objective of the proposed approach?

To integrate NDEA with machine learning and artificial neural networks by combining an upstream prediction/adjustment step with a downstream optimization of economic quantities derived from NDEA solutions.

How does Random Forest regression contribute to the NDEA modelling?

Random Forest regression is applied in regression models to adjust input/output data, sub-process resource allocation preferences, and cost, revenue, and profit targets under the influence of internal and external factors.

What role do artificial neural networks play after NDEA solving?

After resolving the NDEA models, multiple artificial neural network models are used to optimize the calculation of economic quantities of interest obtained from the optimal NDEA solutions, improving robustness of the derived formulas.

Aug 4, 2026

Upload by Valentina