CHASHNIt for enhancing skin disease classification using GAN augmented hybrid model with LIME and SHAP based XAI heatmaps.

Correct categorization of skin diseases is vital for prompt diagnosis. However, obstacles such as imbalance of data and interpretability of deep learning models limit their use in medical settings. To overcome these setbacks, Combined Hybrid Architecture for Scalable High-performance in Neural Iterations or CHASHNIt is proposed, which is an integration of EfficientNetB7, DenseNet201, and InceptionResNetV2 to outperform current models on every ground. GAN-based data augmentation is used to create synthetic images, to ensure that all classes are equally represented. Sophisticated preprocessing methods such as normalization and feature selection improve data quality and model generalization. Explainable AI methods, i.e., SHAP and LIME, enable model decision-making transparent. A rigorous comparative analysis testifies to the excellence of CHASHNIt compared to other benchmark models with 97.8% accuracy, 98.1% precision, 97.5% recall, 97.6% F1 Score and IoU of 92.3%, which exceeds Swin Transformer, ResNet101, InceptionResNetV2, MobileNetV3, EfficientNetB7, DenseNet201, and ConvNeXt models. The model was trained and tested on a 19,500-image dataset of 23 types of skin diseases with 80:20 split for training and testing. An ablation study testifies to the synergy advantage of the hybrid approach. LIME-SHAP heatmaps confirm the model's predictive result. CHASHNIt is an advanced automated skin disease classification framework, attaining a balance between scalability, accuracy, and explainability. Computational complexity is the sole drawback, but future developments will optimize efficiency for low-resource devices.