🚚 AI-Powered Delivery ETA Optimization

Graph-Based Network Intelligence for Logistics

An industry-grade ML + Graph Analytics system for predicting delivery ETAs, detecting bottleneck hubs, and optimizing logistics routes — built with Python, Scikit-learn, NetworkX, and Streamlit.

---

📋 Table of Contents

1. Project Overview
2. System Architecture
3. Features
4. Tech Stack
5. Setup & Installation
6. Usage
7. Dashboard Pages
8. Dataset
9. Model Performance
10. Future Improvements

---

🎯 Project Overview

This project builds an intelligent logistics analytics system that:

• Predicts Delivery ETA using ML regression (Linear, Random Forest, Gradient Boosting)
• Detects Bottleneck Hubs via graph centrality analysis (Betweenness, Closeness, Degree)
• Models the Network as a weighted directed graph using NetworkX
• Optimizes Routes with Dijkstra's shortest-path algorithm (fastest / safest / shortest)
• Generates Business Insights automatically from ML + Graph outputs
• Visualizes Everything via a 9-page Streamlit dashboard

---

🏗️ System Architecture

Raw Logistics Data (5,000 rows)
         │
         ▼
┌─────────────────────────────┐
│   PREPROCESSING PIPELINE    │
│  Dedup → Impute → Engineer  │
│  Encode → Normalize         │
└──────────┬──────────────────┘
           │
    ┌──────┴──────┐
    ▼             ▼
┌────────┐   ┌──────────────┐
│  ML    │   │    GRAPH     │
│Pipeline│   │   Pipeline   │
│        │   │              │
│ Lin Reg│   │ Build Graph  │
│ Rand F │   │ Centrality   │
│ Grad B │   │ Dijkstra     │
└───┬────┘   └──────┬───────┘
    └────────┬───────┘
             ▼
    ┌─────────────────┐
    │  STREAMLIT      │
    │  DASHBOARD      │
    │  + Insights     │
    └─────────────────┘

---

✨ Features

Feature	Description
📦 Data Pipeline	Dedup, imputation, feature engineering, encoding, normalization
🤖 ETA Prediction	3 ML models with R²=0.99 on best model
🕸️ Graph Analytics	20-node, 380-edge directed delivery network
🔍 Bottleneck Detection	Centrality metrics + composite bottleneck score
🗺️ Route Optimization	Fastest / Safest / Shortest path comparison
💡 Business Insights	8+ auto-generated actionable recommendations
📊 Dashboard	9-page interactive Streamlit application
📈 Model Comparison	MAE / RMSE / R² across all 3 models

---

🛠️ Tech Stack

Category	Libraries
Data	pandas, numpy
ML	scikit-learn (LinearReg, RandomForest, GradientBoosting)
Graph	networkx (DiGraph, Dijkstra, Centrality)
Visualization	matplotlib, seaborn
Dashboard	streamlit
Utilities	scipy, pickle, json

---

⚙️ Setup & Installation

# 1. Clone / download the project
cd delivery_eta

# 2. Install dependencies
pip install -r requirements.txt

# 3. Run full pipeline (preprocesses data + trains models)
python main.py

# 4. Launch Streamlit dashboard
streamlit run src/dashboard/app.py

---

🚀 Usage

Run Pipeline Only

python main.py

Launch Dashboard

streamlit run src/dashboard/app.py

Predict ETA Programmatically

from src.models.predict import load_best_model, build_feature_row

model, name, _ = load_best_model()
X = build_feature_row(
    distance=500, traffic=0.6, hub_load=0.5, stops=2,
    weather="Rain", priority="Express", route_type="Highway"
)
eta_hrs = model.predict(X)[0]
print(f"Predicted ETA: {eta_hrs:.2f} hrs")

---

📱 Dashboard Pages

Page	Description
🏠 Home	KPI cards, system architecture, tech stack
📊 Dataset Overview	Sample data, statistics, distribution charts
🔮 ETA Prediction	Interactive form with real-time predictions
🕸️ Graph Network	Network visualization with path highlighting
🔍 Bottleneck Analysis	Centrality table, subgraph, risky routes
🗺️ Route Optimizer	Compare fastest / safest / shortest routes
💡 Business Insights	Auto-generated operational recommendations
📈 Model Performance	Metrics, actual vs predicted, feature importance
🏁 Conclusion	Achievements, future work, deployment notes

---

📊 Dataset

Real-world Delhivery logistics dataset mapped across 20 major Indian logistics hubs.

Column	Description
source_hub	Origin city / warehouse (mapped from raw source name)
destination_hub	Delivery destination (mapped from raw destination name)
route_distance	Distance in km
traffic_level	Traffic congestion factor based on OSRM travel times
weather_condition	Weather condition assigned based on temporal monsoons
num_stops	Intermediate stops
hub_load	Hub capacity utilization load factor
shipment_priority	Economy / Standard / Express / Same-Day
delay_minutes	Actual delay vs OSRM travel time
delivery_time_hrs	Target variable (actual trip travel time in hours)

Engineered features: congestion_score, weather_risk, avg_delay_per_route, est_travel_hrs, is_peak_hour, is_weekend, etc.

---

🤖 Model Performance

Model	MAE (hrs)	RMSE (hrs)	R²
Linear Regression	~8.13	~11.62	~0.904
Random Forest	~3.45	~5.62	~0.978
Gradient Boosting	~2.23	~4.09	~0.988

---

🔭 Future Improvements

• 🌐 Live logistics API integration (Delhivery, FedEx)
• 🗺️ Geo-mapped network with real coordinates (Folium / Kepler.gl)
• ⏱️ Time-series delay forecasting (Prophet / LSTM)
• 🔄 Real-time dynamic rerouting engine
• 🧠 Graph Neural Network embeddings
• ☁️ Cloud deployment (AWS/GCP with Docker)
• 🔔 SLA breach alert system (Email / SMS)

---

📄 License

MIT License — Free for educational and commercial use.

---

Built as an industry-grade capstone project in ML + Graph Analytics.