Classification of EU countries' progress towards sustainable development based on ordinal regression techniques

Research areas:

• Methodology - Machine learning

Year:

2014

Type of Publication:

Article

Keywords:

Sustainable development, European Union, Machine learning, Ordinal regression, Ensemble methods

Authors:

• Pérez-Ortiz, María
• de la Paz Marín, Mónica
• Gutiérrez, Pedro Antonio
• Hervás-Martínez, César

Journal:

Knowledge-Based Systems

Volume:

66

Pages:

178-189

Month:

August

ISSN:

0950-7051

BibTex:

@article{SDviaORKBS, author = "Mar{\'i}a P{\'e}rez-Ortiz and M{\'o}nica de la Paz Mar{\'i}n and Pedro Antonio Guti{\'e}rrez and C{\'e}sar Herv{\'a}s-Mart{\'i}nez", abstract = "Sustainable development (SD) is a major challenge for nations, even more so in the current economic crisis and uncertain environment. Although different indicators, composite indices and rankings to measure and monitor SD advances at the macro level exist, the benefits for stakeholders and policy makers are still limited because of the absence of predictive models (in the sense of models able to classify countries according to their SD advances). To cope with this need, this paper presents a first approximation via machine learning techniques. First, we study the SD stage of the 27 European Union Member States using information from the years 2005-2010 and different major indicators that have been related to SD. A hierarchical clustering analysis is conducted, and the patterns are categorised as advanced, followers, moderate and initiated, according to their progress towards SD. The classification problem is addressed from an ordinal regression point of view because of the inherent order among the categories. To do so, a reformulation of the one-versus-all scheme for ordinal regression problems is used, making use of threshold models (Logistic Regression (LR) and Support Vector Machines in this case) and a new trainable decision rule for probability estimation fusion. The empirical results indicate that the constructed model is able to achieve very promising and competitive performance. Thus, it could be used for monitoring the progress towards SD of the different EU countries, in a manner similar to that used for rankings. Finally, the decomposition method based on LR is used for model interpretation purposes, providing valuable information about the most relevant indicators for ranking the end-point variable.", awards = "JCR(2014): 2.947 Position: 16/123 (Q1) Category: COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE", comments = "JCR(2014): 2.947 Position: 16/123 (Q1) Category: COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE", doi = "10.1016/j.knosys.2014.04.041", issn = "0950-7051", journal = "Knowledge-Based Systems", keywords = "Sustainable development, European Union, Machine learning, Ordinal regression, Ensemble methods", month = "August", note = "JCR(2014): 2.947 Position: 16/123 (Q1) Category: COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE", pages = "178--189", title = "{C}lassification of {EU} countries{\&}#039; progress towards sustainable development based on ordinal regression techniques", url = "http://dx.doi.org/10.1016/j.knosys.2014.04.041", volume = "66", year = "2014", }

Note:

JCR(2014): 2.947 Position: 16/123 (Q1) Category: COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Abstract:

Sustainable development (SD) is a major challenge for nations, even more so in the current economic crisis and uncertain environment. Although different indicators, composite indices and rankings to measure and monitor SD advances at the macro level exist, the benefits for stakeholders and policy makers are still limited because of the absence of predictive models (in the sense of models able to classify countries according to their SD advances). To cope with this need, this paper presents a first approximation via machine learning techniques. First, we study the SD stage of the 27 European Union Member States using information from the years 2005-2010 and different major indicators that have been related to SD. A hierarchical clustering analysis is conducted, and the patterns are categorised as advanced, followers, moderate and initiated, according to their progress towards SD. The classification problem is addressed from an ordinal regression point of view because of the inherent order among the categories. To do so, a reformulation of the one-versus-all scheme for ordinal regression problems is used, making use of threshold models (Logistic Regression (LR) and Support Vector Machines in this case) and a new trainable decision rule for probability estimation fusion. The empirical results indicate that the constructed model is able to achieve very promising and competitive performance. Thus, it could be used for monitoring the progress towards SD of the different EU countries, in a manner similar to that used for rankings. Finally, the decomposition method based on LR is used for model interpretation purposes, providing valuable information about the most relevant indicators for ranking the end-point variable.

Comments:

JCR(2014): 2.947 Position: 16/123 (Q1) Category: COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE