Simone De Bonis
Simone De Bonis
← Back to Projects
Data ScienceMarch 13, 2023

Italian Municipalities Default Prediction

Early warning machine learning system using Support Vector Machines (SVM), Multi-Layer Perceptrons (MLP), and synthetic class balancing (SMOTE & ROSE) to predict municipal financial distress.

#SVM#MLP#SMOTE & ROSE#Credit Risk#Data Science#Machine Learning
View Interactive Dashboard / Repo ↗

Financial distress (dissesto finanziario) in local governments carries severe socio-economic consequences for citizens and public services. In this collaborative university research project, we designed an early warning classification system to detect Italian municipalities at risk of financial default ahead of time.

Methodology & Modeling Pipeline

  1. Feature Engineering & Financial Ratios:
    • Extracted balance-sheet, demographic, debt-sustainability, and revenue-collection indicators across Italian municipalities.
    • Standardized continuous accounting indicators and handled missing structural records.
  2. Addressing Severe Class Imbalance (SMOTE vs. ROSE):
    • Because municipal default is a rare tail event, naive classifiers suffer from near-zero recall on the minority class.
    • We conducted a systematic comparison of SMOTE (Synthetic Minority Over-sampling Technique) and ROSE (Random Over-Sampling Examples) applied strictly inside training folds during cross-validation (see the companion article How to Balance Your Data: SMOTE vs. ROSE).
  3. Supervised Classifiers (SVM & MLP):
    • Trained radial-basis and linear Support Vector Machines (SVM) alongside Multi-Layer Perceptron (MLP) neural networks.
    • Optimized decision thresholds for early-warning sensitivity (Recall / F1 / ROC-AUC) to minimize false negatives on distressed municipalities.
← Back to Projects
Share article: