🩺 Skin Cancer Classification (9-Class ISIC)

---

📌 Project Overview

This project implements a 9-class skin lesion classification system using EfficientNet with transfer learning and Stratified 5-Fold Cross Validation on the ISIC dataset.

The goal is to classify dermoscopic images into clinically relevant categories using deep learning while integrating explainability using Grad-CAM.

---

📂 Dataset

• ISIC Skin Cancer Dataset
• ~2300 Images
• 9 Diagnostic Classes
• Stratified 5-Fold Cross Validation

Classes:

• Actinic Keratosis
• Basal Cell Carcinoma
• Dermatofibroma
• Melanoma
• Nevus
• Pigmented Benign Keratosis
• Seborrheic Keratosis
• Squamous Cell Carcinoma
• Vascular Lesion

---

🧠 Model Architecture

• EfficientNet (ImageNet Pretrained)
• Transfer Learning
• Fine-Tuning Strategy
• Focal Loss
• Data Augmentation
• Grad-CAM Visualization

---

📊 Results

• Average 5-Fold Accuracy: ~55–65%
• Best Validation Accuracy: ~XX%
• Cross-validation improves robustness
• Grad-CAM used for lesion region visualization

---

🔬 Explainability (Grad-CAM)

Grad-CAM is implemented to visualize the regions influencing model decisions, increasing medical interpretability.

---

⚠ Challenges

• Class imbalance
• High inter-class similarity
• Limited dataset size

---

🚀 Future Improvements

• Ensemble models
• Advanced augmentation (MixUp / CutMix)
• Larger dataset training
• Hyperparameter optimization

---

👨‍💻 Author

Arnav Singh
B.Tech Electronics & Computer Science
KIIT University