title	Delivery Time Prediction
emoji	🍕
colorFrom	blue
colorTo	green
sdk	gradio
sdk_version	4.0.0
app_file	ui/app.py
pinned	false

🍕 Delivery Time Prediction

A friendly Gradio demo (XGBoost) that predicts delivery time from order characteristics

• Short: XGBoost regression model that estimates delivery time (minutes) from features like distance, weather, traffic, vehicle type and courier experience.
• Purpose: Give quick, explainable delivery-time estimates to support logistics decisions or power a demo UI on Hugging Face Spaces.

Dataset (quick facts)

• Shape: 1000 rows × 9 columns

• Train / Test split: 795 / 199

• Converted to Pandas for preprocessing and modeling

• Missing values:

  • Weather: 30
  • Traffic_Level: 30
  • Courier_Experience: 30

• Main columns (used in the model)

  • Order_ID
  • Distance_km
  • Weather
  • Traffic_Level
  • Vehicle_Type
  • Preparation_Time_min
  • Courier_Experience_yrs
  • Delivery_Time_min (target)

• Example preview (first 5 rows — one column omitted for brevity):

  • 522, 7.93, Windy, Low, <other_col>, Scooter, 12, 1.0, 43
  • 738, 16.42, Clear, Medium, <other_col>, Bike, 20, 2.0, 84
  • 741, 9.52, Foggy, Low, <other_col>, Scooter, 28, 1.0, 59
  • 661, 7.44, Rainy, Medium, <other_col>, Scooter, 5, 1.0, 37
  • 412, 19.03, Clear, Low, <other_col>, Bike, 16, 5.0, 68

Model & performance

• Algorithm: XGBoost regressor (scikit-learn API)
• CV (k-fold) MAE: 7.15 ± 0.27 minutes
• Final metrics:
  • MAE (train): 3.78 min
  • MAE (test): 6.37 min
  • RMSE (test): 9.55 min
  • R² (test): 0.816

How it works (pipeline)

• Data ingestion → convert to Pandas
• Missing-value handling (imputation / simple strategy for categorical/continuous)
• Categorical encoding for Weather, Traffic_Level, Vehicle_Type
• Train/test split (795 / 199)
• XGBoost training with cross-validation (MAE objective)
• Evaluate on hold-out test set and surface results in the Gradio UI

What you’ll find in this Space

• Interactive Gradio app (app.py) to:
  • Input feature values (distance, weather, traffic, vehicle, prep time, courier experience)
  • See predicted delivery time (minutes)
  • Optionally show confidence or explanation (if SHAP/feature importances added)
• Model served behind the UI (fast inference, suitable for demo / prototyping)

Run locally

• Recommended (example):
  • Create venv: python -m venv .venv && source .venv/bin/activate
  • Install deps: pip install -r requirements.txt
  • Run: python app.py
• Notes:
  • app.py is built with Gradio (sdk_version: 4.0.0)
  • If you include a saved model artifact, name it consistently (e.g., model.pkl) and load with joblib/pickle

Inputs & expected ranges (suggested)

• Distance_km: 0.5 – 20.0
• Weather: categorical — Clear, Rainy, Windy, Foggy, ...
• Traffic_Level: categorical — Low, Medium, High
• Vehicle_Type: categorical — Bike, Scooter, Car, ...
• Preparation_Time_min: integer, typically 5 – 30
• Courier_Experience_yrs: float/integer, typically 0 – 9

Tech stack

• Model: XGBoost
• Serving / UI: Gradio
• Preprocessing & training: pandas, scikit-learn
• Language: Python

Tips & caveats

• Model generalization depends on dataset coverage (geography, vehicle fleets, traffic patterns).
• Missing categorical values (Weather, Traffic_Level, Courier_Experience) were present — choose an appropriate imputation strategy for production.
• Consider recalibrating or retraining with new data periodically to keep predictions accurate.