A Robust Deep CNN Stacked Ensemble Classifier for Skin Disease Diagnosis

Abstract

Skin diseases are more prevalent than other diseases and are caused by bacteria, allergies, and viruses. Advancements in medical technology, including lasers and photonics, enable faster and more precise diagnosis of skin illnesses. However, the expense of such diagnostics is huge, hence the need for automation through computational techniques. Deep learning models are reshaping medical imaging, with a greater emphasis on guaranteeing openness, low cost and confidence in clinical decision‐making processes. This study proposes a stack ensemble learning approach based on three deep convolutional neural networks (CNNS; GGG16, DenseNet121, and Xception) for effective skin disease identification using the HAM10000 and ISIC2020 datasets. By employing advanced augmentation techniques such as rotation, zoom, and mirroring, these models enhance generalization and diagnostic accuracy. The ensemble model significantly outperforms single‐architecture models, with the tri‐hybrid model achieving an accuracy of 94.5% on the International Skin Imaging Collaboration (ISIC) 2020 dataset and 83% on HAM10000. This outstanding performance highlights the suggested approach’s ability to bridge the gap between AI accuracy and skin disease detection using medical imaging. The findings justify the need for well‐defined assessment criteria for a computer‐aided diagnosis approach, as well as the potential inclusion into clinical processes. The is a step forward in developing dependable and automated diagnosis systems that harness the benefits of sophisticated deep learning techniques for smooth adoption into real‐world healthcare settings.