rshewatkar/shipment-delay-predictor

🚚 Shipment Delay Predictor — Logistics ML System

🚀 Live Demo

👉 [Click Here to Try the App](https://rshewatkar-shipment-delay-predictor.hf.space)

Deployed on Hugging Face Spaces with CI/CD (GitHub Actions)

📌 Project Overview

An end-to-end Machine Learning system that predicts shipment delay risk before dispatch.

Built for logistics and courier companies to:

• —Reduce late deliveries
• —Improve customer satisfaction
• —Enable proactive decision-making

⚠️ Key Insight: ~60% of shipments are delayed — this model flags high-risk orders early.

🏠 Dashboard Overview

[image]

🔮 Prediction Flow

[image]

⚠️ Risk Analysis

[image]

🎯 Problem Statement

Late deliveries lead to:

• —Refunds & penalties
• —Customer dissatisfaction
• —Operational inefficiencies

This system predicts delay probability at booking time — enabling early intervention.

📊 Dataset Overview

🔍 Key Insights (EDA)

• —🎯 Discount > 10% → High delay probability
• —⚖️ 2–4 kg weight → Highest delay rate
• —🚢 Ship mode → Most delays
• —📞 Customer calls ≥ 4 → Strong delay signal

⚙️ Feature Engineering

🤖 Model Performance

💡 Business Impact

• —📈 69% of delayed shipments correctly identified
• —🎯 73% precision in delay prediction
• —📉 Potential 15–25% reduction in late deliveries

🖥️ App Features

📁 Project Structure

shipment-delay-predictor/
│
├── app.py                          # Streamlit app (entry point)
├── requirements.txt
├── README.md
│
├── .github/
│   └── workflows/
│       └── sync-to-hub.yml         # CI/CD pipeline (GitHub → Hugging Face)
│
├── data/
│   ├── raw/
│   └── processed/
│
├── notebooks/
│   ├── EDA.ipynb
│   ├── preprocessing.ipynb
│   └── modeling.ipynb
│
├── src/
│   ├── preprocess.py
│   ├── train.py
│   ├── predict.py
│   └── utils.py
│
├── models/                        # Trained ML models
│   ├── best_model.pkl
│   ├── scaler.pkl
│   └── label_encoders.pkl
│
├── reports/                       
│   ├── figures/
│   │   ├──01_target_distribution.png
│   │   ├──02_delay_by_warehouse.png
│   │   ├──03_delay_by_mode.png
│   │   ├──04_discount_effect.png
│   │   ├──05_weight_effect.png
│   │   ├──06_customer_calls.png
│   │   ├──07_correlation_heatmap.png
│   │   ├──08_model_comparison.png
│   │   ├──09_confusion_matrix.png
│   │   ├──10_roc_curve.png
│   │   └──11_feature_importance.png
│   │ 
│   └── screenshots/
│        ├──dashboard-main.png
│        ├──dashboard-result.png
│        └──dashboard-risk.png

⚙️ How to Run Locally

1. Clone the repository

git clone https://github.com/rshewatkar/shipment-delay-predictor.git
cd shipment-delay-predictor

2. Install dependencies

pip install -r requirements.txt

3. Run the app

streamlit run app.py

🔁 ML Pipeline

1. 1.Data Collection
2. 2.Data Preprocessing
3. 3.Feature Engineering
4. 4.Model Training (XGBoost)
5. 5.Model Evaluation
6. 6.Deployment (Streamlit + Hugging Face)

🛠️ Tech Stack

👤 Author

Rahul Shewatkar

• —💼 LinkedIn
• —🐙 GitHub
• —🤗 Hugging Face

📄 License

MIT License