jitennn07/skin-disease-xception-15-classes

Smart Diagnosis: Using CNN to Identify Skin Conditions

This repository contains the official fine-tuned Xception model weights accompanying the research publication:

• Publication: Smart Diagnosis: Using CNN to Identify Skin Conditions (Artificial Intelligence and Sustainable Innovation 2026, Taylor & Francis Group, CRC Press).

Model Details

• Architecture: Fine-tuned Xception-based multi-class Convolutional Neural Network (CNN).
• Task: Multi-class skin disease classification across 15 distinct categories.
• Performance: Achieved 90% validation accuracy (improved from a 88% baseline without 'dual dropout regularization' and 'global average pooling').
• Framework: TensorFlow / Keras.

Training & Methodology Highlights

• Imbalance & Overfitting Mitigation: Successfully resolved a 20:1 class imbalance and closed a 34-point train/validation gap using class weighting, dual dropout regularization, global average pooling, and targeted data augmentation.
• Backbone Benchmarking: Selected Xception following a rigorous benchmark of five backbone architectures (ResNet, VGG, EfficientNet, DenseNet, and Xception) based on superior validation performance across all 15 classes.

How to Load and Use the Model

You can load your saved .keras model directly from Hugging Face using Keras:

import os
os.environ["KERAS_BACKEND"] = "tensorflow"
import keras
from huggingface_hub import hf_hub_download

# Download the .keras file from the Hugging Face repository
model_path = hf_hub_download(
    repo_id="jitennn07/skin-disease-xception-15-classes",
    filename="fine-tuned-xception.keras"
)

# Load the downloaded model locally
model = keras.saving.load_model(model_path)
model.summary()

Clinical Disclaimer

This model is intended strictly for academic research, analytical evaluation, and educational purposes. It is not intended to provide clinical medical diagnoses or replace professional dermatological evaluation.