Functional Information and Dynamic Unconventional Classifiers for Interpretable and Actionable Evaluation in societal impact and policy
Progetto The framework of this project deals with modern statistical methodologies to analyse high dimensional, heterogeneous, and complex data observed over time by considering their evolutive nature. The key challenge it addresses is providing citizens and policy makers with trustable, and useful statistics based on informative data when dealing with decision-making processes, i.e. with elements worthy of their confidence — hence the latin word FIDUCIAE chosen as the acronym of the title. Furthermore, the interest is also in modelling aggregated indicators according to the evolution of longitudinal elementary indicators and their relative importance in explaining social-economic and environmental phenomena. An example is the recent pandemic that caused impacts with relevant social, psychological, economic, environmental, and financial effects. A dashboard capable of summarizing in a timely manner key figures related to the issue at hand may be of great help in visualizing scenarios and parametrizing indicators of the evolutionary impact of any decision-making process and policy. FIDUCIAE focuses on complex evolutionary data. “Complex” refers to the multidimensionality and the support of the data,
e.g. functional data, such as distributional data. Since the special kind of
their supports, it is inappropriate and can prove thoroughly misleading, to apply standard statistical methods designed for observations with more familiar supports. “Evolutionary” refers to the spatial diffusion/dispersion and/or temporal trend over time, like temporal sequences from databases (also by different sources) or streaming data (by sensors), which can be considered either separately or jointly. Functional data represent a typical example of complex evolutionary data, that is high-dimensional data evaluated at any time in the domain. Fostering research for statistical methodologies able to analyse the huge amount of complex data we can collect today is fundamental. The acronym FIDUCIAE emphasises the new concepts and statistical methodologies that will be developed and framed into an innovative paradigm of modelling evolutive phenomena, specifically Functional Information (FI) provided by Functional Data Analysis (FDA) to build up Dynamic Unconventional Classifiers and more informative and useful Composite Indicators (CI), to provide Actionable (such to produce value) knowledge-based Evaluation for stakeholders and decision-makers. Dynamic Unconventional Classifiers are an extension of Supervised Classification (SC) for more stable and readily updated predictions using functional and/or aggregate data as well as other unconventional data such as data stream. The additional purpose is to provide new interpretable, meaningful results and informative Visualization Tools (VT) to overcome black-box algorithms' typical issues.
Therefore, FIDUCIAE relies on FDA, CI, SC, and VT.