Abstract:
"The document discusses the impact of various factors such as sunlight exposure, lifestyle choices, injuries, and infections on skin health, emphasizing the significance of addressing skin infections for both physical and mental well-being. It highlights the challenges faced by previous systems, including high computational demands and limited accuracy, which primarily focused on image recognition.
The research presents a novel hybrid model for skin disease recognition that integrates the EfficientNetV2 architecture, known for its efficiency and accuracy, with an optimized random forest classifier for symptom-based identification. This approach enhances skin disease diagnosis through the incorporation of progressive learning techniques, improving prediction accuracy and robustness.
In the implementation, the EfficientNetV2 model was trained on the Skin Disease Images dataset, yielding optimal performance after 36 epochs, achieving a training accuracy and a training loss of 0.0526. The validation accuracy reached a notable level with a validation loss of 0.5902. Additionally, the Optimized Random Forest Classifier was trained on the Skin Disease Symptoms dataset, achieving a training accuracy of 0.9452 and a testing accuracy of 0.9505. These results indicate significant advancements in model performance for dermatological diagnosis."
