A Comparative Study of VGG16, Support Vector Machine, and Random Forest for Automated Skin Disease Detection Using the DermNet Dataset

This Skin condition classification is crucial for early diagnosis and treatment in dermatology, yet it poses challenges due to the complexity and variability of dermatological images. Traditional classification methods often struggle to balance performance across multiple classes, especially when dealing with rare conditions or those with overlapping features, hindering their reliability in diverse applications. This study aims to evaluate multiple classification models to determine their effectiveness in accurately classifying skin conditions, focusing on identifying the most suitable model for achieving consistent and precise results across all classes. The study compares the performance of three models: Support Vector Machine (SVM), Random Forest, and the VGG16 deep learning model. The models were evaluated based on their accuracy, precision, recall, and consistency across classes. The models were trained and tested on a DermNet image dataset containing samples representing multiple skin conditions, including Lichen Planus and Pityriasis Rubra Pilaris. VGG16 achieved the highest accuracy of 96% by leveraging its ability to analyze image data directly, capturing intricate spatial details crucial for image-based classification tasks. SVM delivered the most balanced performance with 72% accuracy and consistent metrics across all classes, making it ideal for applications requiring uniform class emphasis. Random Forest achieved 70% accuracy and exhibited high recall for Pityriasis Rubra Pilaris, but its lower precision suggested challenges with false positives. In summary, while SVM offers balanced performance and Random Forest may be advantageous for targeted recall, VGG16 demonstrates superior accuracy and precision for dermatological image classification.