<![CDATA[This study utilizes the ResNet50 architecture with transfer learning techniques. The dataset consists of 5,200 facial images categorized into five classes: normal, dry, oily, combination, and sensitive. Data was split with an 80:10:10 ratio for training, validation, and testing. Data augmentation was applied to increase dataset variety. The recommendation system was developed using a content-based filtering approach with rule-based mapping between classification results and product attributes. The ResNet50 model achieved a classification accuracy of 90.4% on test data, with the highest F1-score for the oily class (94.7%) and the lowest for the sensitive class (86.9%). The recommendation system produced a Mean Reciprocal Rank (MRR) of 0.82 and precision@3 of 0.76. User satisfaction testing with 50 participants showed an 84% satisfaction rate. CNN with the ResNet50 architecture is effective for facial skin type classification with high accuracy. The integration of the classification system with content-based recommendation mechanisms successfully provides relevant skincare product recommendations. This system has the potential to become a digital tool that can enhance public skin health literacy and reduce errors in skincare product selection.]]>
Copyrights © 2026
