Enhanced CNN Architecture for Multi-label Skin Condition Classification for Personalized Skincare Recommendations

Skin condition classification is a critical component in developing AI-driven personalized skincare systems. This paper presents a custom convolutional neural network (CNN) based on ResNet50, augmented with attention mechanisms, for multi-label classification of five skin conditions: acne, wrinkles, dry skin, oily skin, and normal skin. The model was trained on a dataset of 880 dermatological facial images, employing condition-specific preprocessing techniques such as Contrast Limited Adaptive Histogram Equalization (CLAHE) and Gaussian smoothing to enhance image quality. Data augmentation strategies were applied to address class imbalance and increase dataset diversity. Using five-fold stratified multi-label cross-validation, the model achieved an average accuracy of 97.25%, precision of 94.44%, recall of 94.72%, F1-score of 94.58%, and AUC of 99.02%. The classification outputs are integrated with a rule-based recommendation engine that incorporates user profiling via a 10-question questionnaire, blending modern dermatology and Ayurvedic practices. This framework demonstrates high performance compared to existing benchmarks, offering a foundation for accessible skincare solutions in diverse populations.