About this project

INCIDB is a multi-source skincare and cosmetics formulation database that transforms raw cosmetic ingredient label text into a normalized, relational dataset. It combines product, brand, and composition data from Open Beauty Facts (ODbL) with ingredient enrichment from the European Commission CosIng inventory, joined on exact canonical INCI name. The dataset contains seven tables: products (19,764 rows), brands (6,412), ingredients (45,327 distinct canonical INCI names), product_ingredients (367,785 ordered composition links preserving label position), ingredient_name_map (77,367 rows mapping raw label tokens to canonical names with resolution method and confidence), fragrance_allergens (274 EU Annex III entries), and regulatory_status (637 EU Annex II–VI rows). Data is delivered as pipe-delimited UTF-8 CSV and Apache Parquet. A key feature is the ingredient_name_map, which records how every raw label token was resolved to a canonical INCI name using explicit methods (exact match, cleaning, typo mapping, percentage stripping, parenthetical stripping, synonym mapping, slash variant handling, splitting, and unresolved flagging). This allows full auditability of the normalization pipeline rather than trusting black-box resolution. Enrichment coverage is reported two ways: by distinct name (12.2% CosIng match) and by label occurrence (85.1% match), reflecting the long-tail nature of ingredient naming. The dataset includes EU fragrance allergen flags (120 names flagged, 47.2% of products contain at least one), EU Annex II–VI regulatory status, comedogenic ratings for 142 ingredients (transcribed from Fulton 1989), and fungal acne trigger flags for 275 ingredients (rule-derived heuristic). A free sample (200 products, 1,123 ingredients) is available, with the complete dataset offered as a one-time $79 purchase. Product data remains ODbL-licensed. Use cases include ingredient scanner apps, formulation and recommender models, retail marketplace enrichment, and regulatory research. A live portal with interactive schema and aggregate statistics is available at incidb.dataengineered.io.