HRP — Stage 06: Date-Specific Room Rate Prediction
Model description
هذا المستودع ينشر النموذج الأصلي المحفوظ من مرحلة HRP Stage 06 بدون إعادة تدريب أو تعديل الأوزان.
- Task: Regression / التنبؤ بسعر الليلة عند تاريخ محدد.
- Selected algorithm:
polynomial_regression_degree_2, sklearn fitted full-feature Pipeline. - Inputs: 37 feature columns (see
models/model_metadata.json). - Output:
suggested_price_per_nightمع العملة USD في نموذج المصدر الاصطناعي. - Caveat: SYNTHETIC / DEMO_ONLY، ولا يغيّر أسعار الفندق تلقائيًا. Currency USD here describes the synthetic training model, not the currency of the operational HRP project.
Recorded original final test metrics
- MAE: 8.6243 USD
- RMSE: 10.7890 USD
- MAPE: 5.37%
- R²: 0.8785
Results were obtained from the original synthetic-data experiment. They are not independently validated on real hotel prices.
Data and evaluation
Training dataset: https://huggingface.co/datasets/AL-mozahem/HRP-Stage06-Synthetic-Room-Rates Original chronological split Train 12,600 / Validation 2,700 / Test 2,700. The public version removes non-feature tracking IDs and preserves 37 feature columns, date and target.
Usage
Install dependencies in a Python environment compatible with scikit-learn 1.8.0:
pip install -r requirements_inference.txt
python example_inference.py
Original model is saved at models/final_model.joblib. Code in src/predict.py is the original inference logic. Only load joblib files from trusted sources.
Limitation / حدود الاستخدام
- Synthetic hotel-like data; not production-validated.
- No direct inference widget is assumed on the Model page; use the Colab notebook for interactive Gradio demonstration.
- Do not use future data as inputs at a past pricing snapshot.
- Recommendations are decision support only, not automated changes.