Skin Disease Prediction using ResNet152V2

A deep learning project that uses ResNet152V2 (transfer learning) to classify various skin diseases from image data. The project includes both a Streamlit-based interactive web app (main.py) and a Jupyter notebook (skin-disease-prediction.ipynb) for training and testing.

🔍 Objective

Early diagnosis of skin conditions can greatly improve treatment outcomes. This project provides an AI-powered tool to classify common skin infections from images using a fine-tuned ResNet152V2 model trained with transfer learning.

---

Kaggle Notebook

---

Key Features

✅ Transfer Learning with ResNet152V2

🖼️ Upload skin images and get predictions instantly

📈 Evaluation via precision, recall, and F1-score

📦 Jupyter notebook for model training & experiments

🌐 Streamlit web app (main.py) for end-user interaction

Model Performance

The model was evaluated on a test set of 229 images across 8 skin disease classes. Below is the classification report:

Class	Precision	Recall	F1-score	Support
BA-cellulitis	0.89	0.94	0.91	33
BA-impetigo	1.00	0.95	0.97	20
FU-athlete-foot	0.96	0.87	0.92	31
FU-nail-fungus	0.94	0.97	0.95	32
FU-ringworm	0.96	1.00	0.98	22
PA-cutaneous-larva-migrans	0.72	0.92	0.81	25
VI-chickenpox	1.00	1.00	1.00	34
VI-shingles	1.00	0.78	0.88	32

---

Overall Metrics:

🔹 Accuracy: 0.93 i.e 93%

🔹 Macro Avg F1-Score: 0.93

🔹 Weighted Avg F1-Score: 0.93

---

📁 Project Structure

Skin-Disease-Prediction/
├── LICENSE
├── README.md
├── main.py                         # Streamlit app
├── skin-disease-prediction.ipynb  # Jupyter notebook (training & analysis)
├── requirements.txt

---

Dataset

Download-->Dataset

---

Model (.h5)

Download-Model

---

Getting Started

1. Clone the Repository

git clone https://github.com/Pranav-Uniyal/Skin-Disease-Prediction.git
cd Skin-Disease-Prediction

2. Install Dependencies

pip install -r requirements.txt

3. Launch Streamlit App

streamlit run main.py

---

Model Details

🔹 Base: ResNet152V2 (pretrained on ImageNet)

🔹 Added Layers: GlobalAveragePooling → Dropout → Dense (Softmax)

🔹 Loss: categorical_crossentropy

🔹 Optimizer: Adam

🔹 Image Input Size: 224x224x3

---

Screenshot

---

Future Scope

Add Grad-CAM visual explanations for prediction insights

Expand to additional skin diseases and rare infections

Deploy live on Streamlit Cloud or Hugging Face Spaces

📃 License

This repository is licensed under the MIT License.

Developed By

Pranav Uniyal Github Kaggle