Project Background
The beauty and skincare industry has grown rapidly, with the number of companies in Indonesia increasing by 21.9% from 2022 to 2023. However, traditional methods for identifying skin types and choosing products are often time-consuming, expensive, and confusing for consumers. Many find it difficult to understand their skin's specific needs and are overwhelmed by the vast number of available products.
GlowGlow was created to address this challenge by offering a sophisticated, user-friendly solution powered by Artificial Intelligence. This initiative also supports Sustainable Development Goal (SDG) 3 for Good Health and Well-being by helping consumers make informed choices, reduce product waste, and minimize negative environmental impacts.
Methodology & Solution
The core of GlowGlow is a recommendation system that uses Convolutional Neural Networks (CNNs) to predict product effectiveness based on skin type. We utilize three powerful architectures: VGG-16 for its strong baseline performance, MobileNetV2 for efficiency on mobile devices, and ResNet-50 for future scalability and handling complex features.
The model was trained on a combined dataset of skin images (categorized as acne, dry, or oily) and product information from CSV files. We performed extensive image preprocessing, including resizing, normalization, and data augmentation, to improve model accuracy and prevent overfitting.
Model Evaluation
The VGG-16 model achieved a final validation **accuracy of 76%** after 50 epochs. The F1-Score was consistent across all classes (0.73-0.79), indicating stable performance. The confusion matrix revealed strong performance for 'dry' skin (81% recall) and 'acne' (76% recall), with some misclassifications that provide clear areas for future improvement.