SkinCare Pro Advisor: AI-Powered Skin Photo Analyzer & Recommender is a comprehensive project aimed at classifying different skin types and conditions using deep learning techniques. This repository contains code for training and evaluating convolutional neural network (CNN) models using transfer learning, along with personalized recommendations based on skin type predictions generated by OpenAI's language model.
A custom dataset was curated from Roboflow's SkinClassification Image Dataset, comprising three folders: Acne, Dry, and Oil. The dataset contains images representing various skin types and conditions, totaling 692 images.
The dataset was split into training and testing sets (80% and 20% respectively). Images were pre-processed, including auto-orientation and resizing.
- Transfer learning was employed using pre-trained models such as MobileNetV2, EfficientNetB0, and VGG16.
- Custom classification heads were added to the base models.
- Data augmentation techniques were applied to increase the diversity of the training data.
Standard evaluation metrics including accuracy, precision, recall, and F1-score were used to assess model performance.
Evaluation metrics for feature extraction models after 30 epochs:
| Model | Accuracy | Precision | Recall | F1-Score |
|---|---|---|---|---|
| MobileNetV2 | 0.94 | 0.94 | 0.94 | 0.94 |
| EfficientNetB0 | 0.43 | 0.43 | 0.43 | 0.43 |
| EfficientNetV2B0 | 0.44 | 0.62 | 0.44 | 0.38 |
| MobileNetV3Small | 0.39 | 0.59 | 0.39 | 0.27 |
| ResNet50 | 0.32 | 0.24 | 0.32 | 0.18 |
| VGG16 | 0.70 | 0.79 | 0.70 | 0.67 |
- MobileNetV2 emerged as the top performer, demonstrating high accuracy levels on both training and validation sets.
- Chatbot integration allowed personalized recommendations based on predicted skin types, enhancing user engagement.
- Data augmentation techniques and transfer learning proved effective for skin type classification.
SkinCare Pro Advisor: AI-Powered Skin Photo Analyzer & Recommender provides a robust framework for skin type classification using deep learning techniques. MobileNetV2 exhibited superior performance, highlighting the effectiveness of transfer learning in dermatological image analysis. Integration with OpenAI's language model enables personalized skincare recommendations, enhancing user experience and engagement.
- Saiwaeo, S., Arwatchananukul, S., Mungmai, L., Preedalikit, W., & Aunsri, N. (2023, November). Human skin type classification using image processing and deep learning approaches. Heliyon, 9, e21176. doi:10.1016/j.heliyon.2023.e21176
- Roboflow. SkinClassification Image Dataset