Ver o conteúdo principal

AICE - Data Science and Over-Indebtedness: Use of Artificial Intelligence Algorithms in Credit Consumption and Indebtedness Conciliation in Portugal, 2019

Topo Projetos

About the Project

Data Science and Over-Indebtedness: Use of Artificial Intelligence Algorithms in Credit Consumption and Indebtedness Conciliation in Portugal.

In the last decades, credit expanded in Portugal, developing different debtors’ classes and social strata. Despite the decline in unemployment, many families continue to face financial difficulties, and a representative part of the Portuguese population still cannot pay their debts.

This project proposes the use of Machine Learning (ML) for developing descriptive and predictive models, to understand the influencing factors of over-indebtedness on Portuguese consumers.

Descriptive models will be obtained using Unsupervised ML algorithms like Self Organizing Maps and Agglomerative Hierarchical Clustering and will be used for establishing consumer clusters and guidelines for over-indebtedness regulation and consumer financial empowerment.

The objectives of the project are:

  • characterize and describe over-indebtedness of Portuguese consumers using unsupervised ML;
  • create reliable supervised ML models to help to predict the factors that influence over-indebtedness;
  • develop interventions to assist in the RAL of consumer debt.

Our team joint researchers with an background in ML, experts of the applicative domain and the Directorate General for Consumer Affairs (DGC). DGC was responsible for contributing to the development, definition, and implementation of consumer protection policy to ensure a high level of protection. DGC included economy, lawyers and psychologists with extensive experience in counseling consumers in financial distress. DGC participated in the transfer of knowledge, providing vast amounts of data of indebted Portuguese consumers, especially the ones who use the RAL centers.

With AICE, we provided reliable predictive models of over-indebtment can serve as a starting point for the development of a set of interventions to assist in the alternative resolution of Portuguese debt disputes. Furthermore, the interpretation of the models can help in making those interventions personalized and effective

Our contribution

NOVA IMS coordinated this project focused on applying machine learning (ML) algorithms to develop both descriptive and predictive models for consumer behavior analysis and financial regulation. Descriptive models were created using unsupervised ML algorithms such as Self-Organizing Maps, Agglomerative Hierarchical Clustering, and Genetic Algorithms. These models were essential for identifying consumer clusters and formulating guidelines for over-indebtedness regulation and consumer financial empowerment.

For predictive modeling, supervised ML algorithms like Decision Trees, Support Vector Machines, various versions of supervised Artificial Neural Networks, and Genetic Programming (GP) were employed. Special attention was given to recent variants of GP, particularly Geometric Semantic GP (GSGP). GSGP demonstrated remarkable success, often surpassing state-of-the-art methods in various application domains. This success was attributed to GSGP's ability to produce a unimodal error surface for any supervised learning problem, regardless of data complexity and dimension. This made GSGP an ideal candidate for addressing the challenges of this project, which involved large and complex datasets.

A thorough comparison of the different algorithms used was essential and was facilitated by Automated ML. This set of techniques automated the ML process, enabling the evaluation of thousands of models with various parameter combinations and feature selection methods.

Interpreting the ML models was crucial for developing interventions that assisted in the Alternative Dispute Resolution (ADR) of consumer debt.


  • Universidade de Lisboa
  • Direção-Geral do Consumidor (DGC), Ministério da Economia

Contribution to the SDGs

  • SDG 1
  • SDG 2
  • SDG 8


  • Funding Programme: FCT, I.P. - Scientific Research and Technological Development Projects in Data Science and Artificial Intelligence in Public Administration
  • Funding to NOVA IMS: 239.243,75 €
  • Duration: 2020 - 2022