Data–driven bayesian networks modelling to support decision–making: application to the context of sustainable development goal 6 on water and sanitation

• Autores: David Requejo Castro
• Directores de la Tesis: Agustí Pérez Foguet (dir. tes.) , Ricard Giné Garriga (codir. tes.)
• Lectura: En la Universitat Politècnica de Catalunya (UPC) ( España ) en 2021
• Idioma: español
• Texto completo no disponible (Saber más ...)
• Resumen

  • We live in a complexand interconnected world which permeates ditterent scales. sectors or decision problems. This fact is acknowledged by the United Nations 2030 Agenda for Sustainable Development, which underscores current global challenges, recognizes their interconnectivity and calls for international action. lt is recognized that the connected nature of the issues we currently face have been tackled by "silo" approaches, separating the complexities ofthe real-world into specialized disciplines. fields of research, institutions and ministries, each one focused on a fraction of the overall truth.

    Similarly, it is widely recognized the need of a major shift in decision-making processes towards more holistic and integrated approaches. Evidence-based decrsion-making involves complexprocesses ofconsidering a wide range of information of different nature. Nowadays, available data can support these processes, but methodologies to effectively integrate these data are lacking.

    With the aim to contribute in this direction, this thesis focuses on the increasing use of Bayesian Networks (BNs) modelling as an approach to accom m odate com plex problem s and to support decis ion-making. Com mon practica em ploys separately expert knowledge and empirical data to build and apply associated models. Des pite of the demonstrated utility of this practica, in an era where the data are bigger, faster and more detailed than even before, there is still room for further exploration. Thus, this dissertation proposes a data-driven Bayesian Networks approach to combine expert opinion and quantitative data to support informad decision-making.

    We propose two systematic methods to this end. First. we use our approach to replicate composite indicators (Cl)-based conceptual frameworks, which represent expert knowledge. through the use of structure learning algorithms, which characterizes this data-driven Bayesian Networks approach. Second, we use our approach to identify interlinkages associated with a complex context, coupled with a statistical technique (i.e. bootstrapping) to reduce results uncertainty and with a comprehensive result robustness analysis (i.e. expert knowledge).

    For testing and validating the proposed approach, this thesis takes the Sustainable Development Goal 6 embedded on the 2030 Agenda as a reference point, with particular attention to the water, sanitation and hygier:ie sector.

    Our results emphasize the likely utility of the data-driven Bayesian Networks approach adopted. First. it allows the integration of both expert knowledge and data availability when dealing with BNs modelling, and it accurately replicates (Cl)-based conceptual frameworks. As added values, this combination improves model inference capacity, it reduces and quantifies the key variables that explµin a pre-defined objective variable (implying important advantages in data updating), and it identifies the interlinkages among the variables considerad (which might enhance more integrated actions). Second, the approach adopted is useful to accommodate a thorough analysis and interpretation of the complexities and interdependencias of any context at hand. As added values, interlinkages identification is spurred on by the available data and this identification makes the approach more suitable than the use of composite indicators. Third, the systematic nature of the methodological contributions associated with the proposed approach can be adapted to different complexproblems. Thus, it might expand and deepen the knowledge about the validity, reliability and accuracy of using BNs modelling.