Online comment-based prediction of cosmetic ingredient's sensory irritation using gradient boosting algorithm

Area of Science:

• Dermatology and Cosmetic Science
• Computational Biology
• Data Science

Background:

• Approximately 40% of the global population experiences "sensitive skin."
• Misuse of cosmetic products is a primary trigger for sensitive skin reactions.
• Current in vitro and in vivo testing methods for cosmetic ingredients have limitations in assessing sensory irritation and real-world applicability.

Purpose of the Study:

• To develop a predictive model for sensory irritation caused by cosmetic ingredients.
• To integrate diverse data sources including skin types, sensory responses, and cosmetic ingredients.
• To overcome the limitations of existing in vitro and in vivo testing methodologies.

Main Methods:

• Online user comments were preprocessed to create a structured, multi-dimensional dataset.
• A gradient boosting regression model was developed to predict sensory response.
• Model predictions were validated using in vivo testing and compared against state-of-the-art methods.

Main Results:

• The model successfully predicted sensory responses for 16 skin types across various ingredients (R = 0.71, P < 10^-10).
• Validation via in vivo studies demonstrated high performance metrics: 75% specificity, 88.9% sensitivity, and 82.4% accuracy.
• The study utilized 46,007 cleaned samples for model development and validation.

Conclusions:

• The developed gradient boosting algorithm variant offers an effective solution for understanding cosmetic ingredient sensory irritation.
• This data-driven approach enhances the assessment of cosmetic ingredient safety and efficacy.
• The findings pave the way for more reliable prediction of skin reactions to cosmetic products.