AI-Driven Ingredient Association Analysis and Recommendation System for Skincare E-Commerce Platforms

Abstract

This research aims to analyze relationships among ingredients in skincare products and to develop an automatic recommendation system for e-commerce platforms using data mining and artificial intelligence (AI) techniques. Association Rule Mining (Apriori Algorithm) was applied to a dataset of 2,441 skincare products to identify significant co-occurrence patterns, such as niacinamide → glycerin (Confidence: 92.3%, Lift: 1.29) and caffeine & sodium hyaluronate → butylene glycol (Confidence: 93.7%, Lift: 2.10). K-Means clustering categorized products based on ingredient similarity into four groups: moisturizing, sensitive skin care, acne treatment, and general/natural care. The clustering performance—measured by a Sum of Squared Error (SSE) of 52.67 and a Percentage Error of 12.8%—indicated acceptable accuracy. The recommendation system integrates content-based and collaborative filtering with machine learning techniques, including Logistic Regression and Decision Trees. Trained using 10-fold cross-validation, the models achieved up to 87.2% accuracy, an average F1-score of 0.876, and an AUC of 0.92. The system supports three key functions: (1) recommending ingredient combinations, (2) suggesting substitutes for out-of-stock products, and (3) personalizing recommendations based on skin type. Real-time API deployment and dynamic model updating are also supported. The results demonstrate the practical integration of AI and ingredient-level analysis for intelligent product recommendations in skincare e-commerce.