@inproceedings{villate-lemus-2026-lafed,
    title = "{LAFED} at {S}em{E}val-2026 Task 13: Language-Agnostic Feature Engineering for Cross-Lingual {AI}-Generated Code Detection",
    author = "Villate Lemus, Juan",
    editor = "Kochmar, Ekaterina  and
      Ghosh, Debanjan  and
      North, Kai  and
      Komachi, Mamoru",
    booktitle = "Proceedings of the 20th {I}nternational {W}orkshop on {S}emantic {E}valuation (2026)",
    month = jul,
    year = "2026",
    address = "San Diego, California, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl-workshops/2026.semeval-1.9/",
    pages = "59--64",
    ISBN = "979-8-89176-414-9",
    abstract = "Robust detection of AI-generated source code across programming languages remains challenging due to language-specific cues and train{--}test distribution shifts. We present LAFED (Language-Agnostic Feature Engineering Detector), a feature-engineering approach trained on {\{}Python, Java, C++{\}} and evaluated on a multilingual test set that includes unseen languages {\{}C, C{\#}, Go, JavaScript, PHP{\}}. LAFED combines (i) structural skeletal features (indentation, control-flow density, and approximations of McCabe/Halstead complexity), (ii) character and whitespace statistics inspired by stylometry, and (iii) micro-style patterns (operator spacing, blank lines, indentation consistency). Using XGBoost (Chen and Guestrin, 2016) with Optuna hyperparameter search (Akiba et al., 2019), our best model achieves macro-F1=0.7570 on a 1,000-sample test set; the official submission obtains macro-F1=0.75209 (5th place in Subtask A). Per-language analysis shows strong transfer to C{\#} (0.7753) and JavaScript (0.7683), but weaker performance on Go (0.6400) and PHP (0.5238)."
}