AL-mozahem/HRP-Stage07-Occupancy-Forecast

HRP Stage 07 — Next 7-Day Room Occupancy Forecast

SYNTHETIC / DEMO_ONLY — NOT PRODUCTION VALIDATED

This model predicts the realized occupancy fraction (0–1) for one hotel room type over the next seven days from features known at snapshot time.

Model

• Original selected estimator: CatBoost regression (depth 6, 120 iterations, learning rate 0.055, L2 leaf reg 5).
• Original serialized artifacts: models/final_model.joblib (and native models/final_model.cbm), models/catboost_numeric_medians.json, models/model_metadata.json.
• Input: all 47 feature names defined in metadata; model pipeline returns predicted occupancy and an advisory expected additional occupancy rate.
• Dataset: https://huggingface.co/datasets/AL-mozahem/HRP-Stage07-Synthetic-Occupancy

Final synthetic test results

Measure	Value
MAE	0.02484 (2.48 percentage points)
RMSE	0.03373 (3.37 percentage points)
R²	0.88062
MAPE	3.24%

These metrics are from the synthetic temporal holdout only and are not evidence of production performance.

Try it from a local folder or Colab

pip install -r requirements_inference.txt
python example_inference.py

Use the exact provided sample payload in examples/sample_cases.json; feature names and types should match the metadata. example_inference.py invokes original src/predict.py, loading the saved model without retraining. A companion Gradio demo is available in the separately shared Colab notebook.

Original provenance and safeguards

• Origin: HRP Machine Learning Stage 07 complete package.
• All features must have been known at snapshot_at for each prediction; do not use future bookings to make predictors.
• Decision support only. No automatic changes to prices, inventory, or bookings.
• Real use requires authentic historical snapshots, realized 7-day outcomes, temporal evaluation and drift checks.