sharvari2/Hallucination-detection-LLM

Repository for detecting hallucinations in LLM-generated answers. It builds a 5,000-sample SQuAD v1.1 dataset, generates answers, labels them, extracts features and trains nine models to classify hallucinations. Intended for ML/NLP practitioners who need lightweight detectors or baseline comparisons. Key results: a 1‑D CNN over concatenated question+context+generated answer achieves highest test accuracy (0.976). Simple classical baselines (logistic regression, random forest) using three handcrafted features—length difference, TF‑IDF cosine similarity, and a context‑match flag—are competitive (~0.96). Includes two notebooks (data build and full pipeline), processed datasets, model comparison plot, and uses Python, PyTorch/TensorFlow, Hugging Face, scikit‑learn. Requires Python 3.10+; GPU optional.

AI-related keywords found in this repo's description, topics, or README summary — grouped by category. Each badge links to the corresponding ranking detail page.