An Exploratory DEA and Machine Learning Framework for the Evaluation and Analysis of Sustainability Composite Indicators in the EU

Abstract

One method that has been proposed for the measurement of sustainability is Data Envelopment Analysis (DEA). Despite its advantages, the method has limitations: First, the efficiency of Decision-Making Units is calculated with weights that are favorable to themselves, which might be unrealistic, and second, it cannot account for different perceptions of sustainability; since there is not an established and unified definition, each analyst can use different data and variations that produce different results. The purpose of the current paper is twofold: (a) to propose an alternative, multi-dimensional DEA model that handles weight flexibility using a different metric (an alternative optimization criterion) and (b) the inclusion of a computational stage that attempts to incorporate different perceptions in the measurement of sustainability and integrates machine learning to explore country sustainability composite indices under different perceptions and assumptions. This approach offers insights in areas such as feature selection and increases the trust in the results by exploiting an inclusive approach to the calculations. The method is used to calculate the sustainability of the 28 EU countries.