Published in the Journal of Open Access to Law (JOAL), hosted at Cornell Law School.
DOI: https://doi.org/10.63567/34drrn20 · PDF: https://ojs.law.cornell.edu/index.php/joal/article/download/61/68

Published my first paper about Trustworthy Predictive Justice

Thesis: predictive justice is only credible if both the data and the code are open to scrutiny. Black‑box “predictions” don’t meet professional standards for lawyers or judges.

In a large‑scale experiment on more than 18,000 decisions from French administrative courts of appeal (2012–2015), a supervised ML model (XGBoost) predicts outcomes in the immigration‑removal docket with 87.2% accuracy and remains interpretable enough for legal professionals to audit. Among the practical signals: noting that a litigant has legal aid increases the likelihood of annulment, while the identity of the presiding judge adds little predictive power. The project releases the model, code, and annotated dataset openly so others can verify, reuse, and critique.

👉 Read the paper (French): L’open data et l’open source, des soutiens nécessaires à une justice prédictive fiable?