Proposed Predictive & Analytical Framework
To address the convergence of population decline and economic fragility, this project implements a municipal-level predictive data science framework. The approach moves beyond descriptive dashboards to forecast trajectories and quantify the relative importance of fertility, age structure, and hazard exposure.
1. Population Forecasting
We employ regularized regression (Ridge/Lasso) and tree-based models (Random Forest, Gradient Boosting) to forecast percent population change. Unlike standard cohort models, this approach integrates demographic momentum features (median age, birth rates) with economic covariates and hazard data to separate structural decline from shock-related migration.
2. Driver Analysis & Explainability
To open the "black box" of prediction, we apply SHAP values and permutation importance. This allows us to rank drivers of change, explicitly comparing the influence of structural factors (low fertility) against policy-amenable factors (employment density, housing vacancy).
3. Municipal Typologies
Using K-means clustering, we identify distinct municipal profiles that share trajectories. This helps classify municipalities into actionable groups, such as "Low-Fertility/High-Migration" or "Hazard-Exposed/Aging," facilitating targeted rather than blanket policy interventions.
4. Risk Scoring & Scenario Analysis
The model outputs feed into a Municipal Demographic Vulnerability Score. We simulate simplified scenarios (e.g., improved employment vs. stagnation) to assess potential resilience even under the constraints of low fertility and demographic momentum.
