Ensemble of classical ML models achieves 99.4% F1 score for AI detection
November 13, 2026
The Unibuc team achieved a 99.4% F1 score in the Voight-Kampff AI detection task using a majority vote ensemble of LightGBM, XGBoost, Logistic Regression, and SVM. This ensemble approach performed comparably to fine-tuning LLMs with a classification head.
HOW THIS AFFECTS YOU
●
builderIf you need to detect AI text, a lightweight ensemble of XGBoost or SVM might be more efficient than a dedicated LLM.
●
researcherYou can achieve high F1 scores for AI detection without heavy fine-tuning by using classical ML ensembles on LLM embeddings.