Transfer learning approaches for EfficientNetV2 B0 and ImageNet skin cancer classification in convolution neural network 

Dataset Link
The data and code are available at Mendeley:
- G, Karthiha (2025), “Dataset and Code for Transfer Learning Approaches for EfficientNetV2 B0 and ImageNet skin cancer classification in convolutional neural network”, Mendeley Data, V2, doi: 10.17632/j7tjmjzks5.2
The third-party dataset used is available at the following link:
- https://challenge2020.isic-archive.com/

Introduction
This project focuses on skin cancer classification, specifically identifying melanoma and benign skin lesions using transfer learning with EfficientNet V2 and ImageNet. The model is trained on the ISIC2020 dataset, incorporating data augmentation and CNN-based architectures to enhance accuracy.

Packages
The code incorporates data preprocessing, machine learning, deep learning, and visualization using various Python libraries. It combines frameworks such as scikit-learn, PyTorch, TensorFlow, and others for different stages of analysis and modeling.

Key Features
- Uses EfficientNet V2 B0 and ImageNet pre-trained models for classification.  
- Implements data augmentation for improved generalization.  
- Incorporates CNN model training and evaluation with TensorFlow and Keras.  
- Compares model performance using confusion matrices and ROC curves.  
- Achieves 99.2% accuracy in melanoma detection.  

Installation & Dependencies  
Ensure you have the following dependencies installed:  
pip install tensorflow tensorflow-hub numpy pandas matplotlib seaborn scikit-learn scikit-plot

How the Model Works  
1. Dataset Preparation
- Loads and preprocesses the ISIC dataset.  

2. Data Augmentation
- Enhances training images to improve model robustness.  

3. Model Training
 - CNN-based approach with multiple convolutional layers.  
 - EfficientNet V2 Transfer Learning for better feature extraction.  

4. Evaluation & Visualization
- Uses confusion matrices, accuracy curves, and ROC curves for performance analysis.