Beyond Sociodemographics: Designing and Implementing a Machine Learning Pipeline for Contextual Risk Analysis of Mafia Infiltration in Italian Municipalities

The thesis addresses the threat of Mafia infiltration into Italian local administrations by developing an analytical pipeline for socio-demographic risk assessment. The methodological architecture proceeds in four sequential steps: data integration, preliminary variable analysis, building a robust predictive Machine Learning model, and translating the numerical risk into an interpretable textual explanation using an LLM. The project’s core objective is to transform the technical output into an effective decision-support tool for public sector professionals.