OPTIMIZING DEEP LEARNING MODELS FOR LARGE-SCALE, MULTI-CLASS SKIN DISEASE CLASSIFICATION IN CLINICAL DERMATOLOGY

Abstract

Based on deep learning skin disease classification has progress a key tool To improve dermatological diagnosis, especially on a large scale, multi- class scenarios. By using a balanced dataset more than 260, 000 dermatoscopic images from 35 different skin disease classes which was taken from Kaggle, This study investigates the capabilities of top convolutional neural network( CNN) architectures. Appreciate the model InceptionResNetV2, sectionv3, ResNet50, Efficient NetB4, and MobileNetV2 I am evaluated the study, side by side ensemble strategies Favor EfficientNetB3 in conjunction with ResNet50 and MobileNetV2. Keeping exercise hours manageable, ResNet50 prepared the best validation accuracy K 97.91% Watch carefully among them InceptionResNetV2 and InceptionV3 on 93.94% And 93.02%, In addition, respectively ensemble models demonstrated a high capacity To generalize. These findings Illustrate how vital it is to distribute AI Finding a balance between accuracy, scalability and computational efficiency AI models I clinical dermatology, Especially in a resource- constrained environment. By encouraging early and effectively detection of skin disorders, This study demonstrates the authority of AI diagnostic systems.