Ms21063/Tourism_Project

Visit with Us — Wellness Tourism Purchase Prediction

End-to-end MLOps project using scikit-learn/XGBoost, MLflow, Hugging Face Hub, GitHub Actions and a public Hugging Face frontend.

Important deployment note

Hugging Face's current Space creation UI may show Docker as paid for free personal accounts. The project therefore includes both:

• deployment/streamlit_app.py — the Streamlit frontend required by the assignment and the Docker configuration.
• deployment/app.py — a Gradio frontend used for the free Hugging Face Space deployment when Docker is unavailable.

The assignment's required Dockerfile remains in deployment/Dockerfile. If Docker Spaces are enabled on your HF account, the Dockerfile can be used directly. Otherwise, use the included Gradio Space deployment.

Required GitHub secrets

• HF_TOKEN
• HF_DATASET_REPO — e.g. Ms21063/tourism-dataset
• HF_MODEL_REPO — e.g. Ms21063/tourism-purchase-model
• HF_SPACE_REPO — e.g. Ms21063/tourism-wellness-predictor

Local run

pip install -r deployment/requirements.txt
python model_building/data_prep.py
python model_building/train.py
cd deployment
streamlit run streamlit_app.py

For the free Hugging Face frontend:

cd deployment/hf_space
pip install -r requirements.txt
python app.py

Executed benchmark

On the supplied dataset (4,128 rows), the executed notebook selected Random Forest by test ROC-AUC (0.9681). Test accuracy was 0.8991, precision 0.9375, recall 0.5097 and F1 0.6611.