Des méthodes ouvertes, des preuves vérifiables

R package for probabilistic and deterministic record linkage — deduplication and entity resolution across data sources, with comparison, classification, and evaluation tools.

Robust synthetic microdata generation when the training data are contaminated — a drop-in alternative to synthpop using MM-estimators, weighted logistic regression, and tree-based robust methods so outliers and coding errors don't propagate into the synthetic output.

R package for statistical disclosure control of microdata — risk estimation, anonymization methods, and utility measurement. Peer-reviewed in the Journal of Statistical Software.

Simulation of synthetic populations from survey and census data.

Design-aware, robust synthetic data generation — the successor to robSynth. Replaces the conditional models in sequential synthesis with MM-estimators and Huber-weighted logistic regression, keeping the synthetic data close to the clean data-generating process even when the source is contaminated.