Hybridized CNN-Densenet Model for Dermatological Classification of Histopathological Skin Images

This research paper presents a deep neural network model for the classification and prediction of skin diseases from dermatoscopic images. The model integrates a traditional Convolutional Neural Network with a Pre-trained Densenet model, achieving state-of-the-art performance on the HAM10000 dataset. The methodology involved rigorous data preprocessing, including data cleaning, exploratory data analysis, data splitting, normalization, and label encoding. Techniques such as Model Hybridization, Batch Normalization, Max Pooling, and Data Fitting were employed to optimize the model architecture and data fitting. Initial iterations of the CNN model achieved an accuracy of 76.22% on Test Data and 75.69% on Validation Data. To improve, the model was hybridized with DenseNet architecture and hyperparameter tuning was employed. The model demonstrated an impressive accuracy of approximately 89% on the HAM10000 dataset. This model offers customization potential for more nuanced clinical uses and presents a significant advancement in automating skin disease diagnosis.