🔍 What Was Tested
A prototype system for diagnosing electoral fraud using only vote counts was developed and evaluated.
🧠 How the Classifier Was Built
- Synthetic data were generated to develop and train a fraud-detection prototype.
- A naive Bayes classifier served as the learning algorithm.
- Digital feature analysis identified which vote-count features are most informative about class distinctions.
📊 What Data Were Used to Evaluate It
- Authentic district-level vote counts from a novel dataset covering the province of Buenos Aires (Argentina) between 1931 and 1941—a period with a checkered history of fraud.
✅ Key Findings
- Elections that historians consider irregular are unambiguously classified as fraudulent by the classifier; elections considered legitimate are classified as clean.
- These results corroborate the validity of the synthetic-data training approach.
- More broadly, the findings demonstrate the feasibility of generating and using synthetic data to train and test an electoral-fraud detection system.
📌 Why It Matters
- Provides a practical, reproducible method to detect fraud from vote counts alone.
- Shows synthetic-data training can overcome the scarcity of labeled historical fraud cases and enable systematic testing of detection tools.
