Integrated Deep Learning and Machine Learning Framework: A Hybrid Approach for Skin Disease Classification

Seeking the right treatment and support is impor-tant to maintain physical and mental health, and it depends largely on the medical science. Skin disease is not only about physical health as it can seriously affect a person's mental well-being if they're not treated properly and if not treated on time. Every sector of medicine relies on images. Analyzing those images can be time-consuming and sometimes it becomes difficult to figure out where the problem is with bare eyes. That's why we need computerized diagnostic systems to ensure consistency and standardization of dermoscopy image analysis. This technology helps to ensure more accurate and consistent results. The study proposes a hybrid model, combining concatenated CNN models with an SVM classifier and a k-NN classifier in the output layer to classify skin disease from the images. A CNN model is trained and combined with a modified pre-trained model. Then, features are extracted from the combined model and fed into SVM and k-NN classifiers for classification. The proposed model performs better compared to the adapted CNN model on our dataset. Using only CNN, the model achieved an accuracy of 88.78%. When we combined CNN with VGGI6, the accuracy improved to 95.17%. Further improvements were seen with our hybrid models: the CNN + VGG16 + k-NN combination achieved 97.95% accuracy, while the CNN + VGG 16 + SVM model achieved the highest accuracy of 98.38 %. These results clearly demonstrate the strength of our proposed framework and show that combining traditional machine learning with deep learning techniques can significantly enhance the accuracy of the classification of skin diseases.