A Study on “Clean" in Beauty: A Machine LearningApproach to Ingredient Transparency and ConsumerTrust

Abstract

The burgeoning "clean beauty" market, while driven by consumer demand for safer and more sustainable products, is plagued by ambiguous definitions and the pervasive challenge of "greenwashing". This ambiguity hinders informed consumer choices and complicates brand authenticity. This study addresses these complexities by developing a novel machine learning (ML) framework designed to objectively analyze cosmetic ingredient lists, classify products based on their "cleanliness" profile, and identify key ingredient attributes that correlate with consumer perception and scientific safety. Leveraging extensive datasets comprising ingredient databases, scientific literature on toxicology and allergens, and public regulatory guidelines, our approach utilized natural language processing (NLP) for efficient ingredient parsing and feature extraction. Various supervised learning models (e.g., Random Forest, Gradient Boosting) were then trained to predict a multi-faceted "cleanliness score" or categorical classification. The results demonstrate superior accuracy in objectively categorizing products, not only distinguishing between "clean" and "non- clean" formulations but also identifying commonly used "cleanwashing" ingredients. Furthermore, the ML models uncovered latent correlations between ingredient profiles and consumer-perceived "cleanliness" derived from sentiment analysis of product reviews, highlighting a significant alignment potential. This research offers a robust, data-driven approach to demystify clean beauty, empowering consumers with greater transparency, guiding brands toward authentic product development, and informing regulatory efforts to standardize health and environmental claims in the cosmetic industry.