Deep Learning in Dermatology: Exploring Convnext Model Hierarchies and Ensembles for Enhanced Diagnostic Precision

Skin cancer is one of the most widespread and life-threatening malignancies in the world that requires early and proper accurate ‎diagnosis to be efficiently solved. This research article examines the performance of different ConvNeXt models- ConvNeXt-Tiny ‎to ConvNeXt-XLarge architectures, and also their ensemble representations in malignant categories of dermoscopic images. To guarantee the ‎balanced representations of classes, using the HAM10000 dataset, a stratified 80-20 train-validation configuration was used to train and real-‎ize the models. Using the indicators of performance in terms of accuracy, precision, recall, F1-score, and confusion matrices, the analysis ‎has shown that middle-ground models and layers, like ConvNeXt-Small combination of performance and efficiency. Ensemble learning also ‎contributes to diagnostic robustness, with ConvNeXt-Tiny, Small, and Base models combined, achieving the greatest validation accuracy of ‎‎92.42%. Comparative analysis shows that, although adding depth leads to marginal benefits in the model, the depth poses a risk of overfit-‎ting and the cost of computation further‎.