Development of a Deep Learning-Based Cosmetic Skin Compatibility Evaluation System

This study develops and validates an automated skin compatibility assessment system for cosmetic products using deep learning-based image recognition technology. The system employs the Gemini 2.5 model designed to perform expert-level analysis, integrating users’ skin type, skin concerns, sensitivity level, and ingredients to avoid with automatically recognized ingredient label images to evaluate product compatibility. The system generates comprehensive reports including numerical scores on a 0-100 scale and detailed ingredient-level assessments. Users can intuitively determine product suitability and clearly understand ingredient-specific effects aligned with their individual skin characteristics without professional consultation. System validation was conducted with 20 university students, revealing no statistically significant difference between system-generated scores and participants’ subjective evaluations, confirming high consistency. This demonstrates that the system achieves reliable compatibility prediction aligned with actual user perception. This research establishes a quantitative framework for skin compatibility assessment through integration of automated ingredient label image recognition and personalized skin data analysis. By utilizing prompt engineering to deliver expert-level judgment in visual and descriptive formats, this study demonstrates the feasibility of AI-driven decision-support systems that democratize access to professional-grade cosmetic analysis.