Logistics Network Intelligence System

Overview

This project builds an end-to-end logistics network intelligence framework for analyzing shipment movement, identifying bottlenecks, predicting ETA delays, and optimizing transportation mode decisions between Full Truck Load (FTL) and Carting operations.

The pipeline combines:

• Exploratory Data Analysis (EDA)
• Graph-based logistics network construction
• Bottleneck and corridor analysis
• ETA prediction using ML models
• Graph embeddings using Node2Vec

The system transforms raw shipment movement data into operational intelligence for network strategy teams.

View Report Here: https://drive.google.com/file/d/1V_9cDIRs3gj0Edc0-NrF5iAJBZSEyPs2/view?usp=sharing

---

Problem Statement

Modern logistics networks face:

• Hub congestion
• Unpredictable delivery delays
• Corridor inefficiencies
• Poor transport mode allocation
• Revenue leakage due to SLA failures

The objective of this project is to:

1. Identify high-friction hubs and corridors
2. Predict shipment ETA accurately
3. Quantify operational bottlenecks
4. Recommend FTL vs Carting decisions
5. Estimate business and revenue impact

---

Repository Structure

├── notebooks/
  ├── 01_eda.ipynb
  ├── 02_graph_construction.ipynb
  ├── 03_bottleneck_analysis.ipynb
  ├── 04_eta_prediction.ipynb
  ├── 05_graph_embeddings.ipynb
├── data/
└── report/

---

Workflow

1. Exploratory Data Analysis

Notebook: 01_eda.ipynb

Performed:

• Null value analysis
• Delay ratio calculation
• Segment-wise delay analysis
• Time distribution analysis
• Shipment duration trends

Key engineered features:

• delay_ratio = actual_time / osrm_time
• segment_delay_ratio
• Hub-level delay aggregation

Insights:

• Significant variance exists between predicted OSRM travel time and actual shipment movement.
• Certain hubs consistently contribute to cascading delays.
• Long-haul corridors show nonlinear delay amplification.

---

2. Graph Construction

Notebook: 02_graph_construction.ipynb

The logistics network is modeled as a directed weighted graph.

Nodes

• Logistics hubs
• Source centers
• Destination centers

Edges

• Shipment movement corridors

Edge Weights

• Median delay ratio
• Corridor traffic frequency

Graph Features

• Degree centrality
• Betweenness centrality
• Connectivity analysis

Why graph modeling? Traditional tabular analytics fail to capture network dependencies. Graphs enable:

• Bottleneck identification
• Route criticality estimation
• Corridor dependency analysis
• Network flow optimization

---

3. Bottleneck Analysis

Notebook: 03_bottleneck_analysis.ipynb

This stage identifies:

• High-delay corridors
• Congested hubs
• Critical transit dependencies

Methodology

1. Aggregate median delay ratio per corridor
2. Rank corridors by delay intensity
3. Analyze centrality metrics
4. Identify hubs with high transit dependency

Key Operational Signals

• Hubs with high betweenness centrality are operationally critical.
• Small disruptions at these hubs create cascading SLA failures.
• Certain corridors exhibit structurally high delay ratios and require intervention.

---

4. ETA Prediction

Notebook: 04_eta_prediction.ipynb

Machine learning models are used to predict shipment ETA.

Features Used

• Source center
• Destination center
• OSRM time
• Segment travel features
• Centrality features
• Corridor statistics

Modeling

Potential models used:

• LightGBM
• CatBoost
• XGBoost
• Ensemble learning

Objective

Minimize ETA prediction error and improve shipment planning reliability.

Business Value

• Better customer SLA adherence
• Dynamic dispatch optimization
• Reduced operational uncertainty

---

5. Graph Embeddings

Notebook: 05_graph_embeddings.ipynb

Node2Vec is used to learn dense vector embeddings for hubs.

Why Node2Vec?

Graph embeddings capture:

• Structural similarity
• Traffic behavior similarity
• Transit role similarity

Benefits

• Better ML feature representation
• Improved ETA prediction
• Corridor similarity clustering
• Hidden bottleneck discovery

---

FTL vs Carting Decision Framework

Objective

Determine the optimal transportation mode based on:

• Shipment volume
• Corridor stability
• Delay risk
• Cost efficiency
• Delivery urgency

---

Definitions

FTL (Full Truck Load)

Dedicated vehicle assigned to a shipment.

Advantages:

• Faster transit
• Lower handling risk
• Better SLA consistency
• Reduced touchpoints

Disadvantages:

• Higher cost for low utilization
• Inefficient for fragmented loads

---

Carting

Shared shipment consolidation model.

Advantages:

• Lower cost
• Better utilization
• Efficient for low-volume shipments

Disadvantages:

• Higher transit uncertainty
• Multiple handling points
• Greater delay probability

---

Decision Framework

Parameter	FTL Preferred	Carting Preferred
Shipment Volume	High	Low
Corridor Delay Variance	Low	Moderate
SLA Criticality	High	Medium
Transit Frequency	Stable	Variable
Shipment Value	High	Low
Cost Sensitivity	Medium	High
Delivery Urgency	High	Low

---

Supporting Analysis

When to Use FTL

Recommended for:

• High-frequency corridors
• Enterprise clients
• Time-sensitive shipments
• Stable long-haul routes

Expected Benefits:

• 15–25% lower delay probability
• Better ETA reliability
• Reduced re-routing risk

---

When to Use Carting

Recommended for:

• Low shipment density corridors
• Non-critical deliveries
• Rural or fragmented networks

Expected Benefits:

• Better fleet utilization
• Lower operating cost
• Higher network flexibility

---

Corridor Intelligence Strategy

High-Risk Corridors

Characteristics:

• High delay ratio
• Large variance
• Multi-hop dependency

Recommended Action:

• Shift high-priority loads to FTL
• Add transit buffers
• Introduce dynamic routing

---

Stable Corridors

Characteristics:

• Predictable transit time
• Lower congestion
• Balanced throughput

Recommended Action:

• Continue carting optimization
• Improve consolidation efficiency

---

Revenue Impact Estimation

Operational improvements can generate revenue impact through:

1. Reduced SLA penalties
2. Improved customer retention
3. Higher shipment throughput
4. Lower idle fleet cost
5. Better route planning

Estimated Impact Areas:

• 8–12% reduction in delay-related penalties
• 10–18% operational efficiency gain
• Improved fleet utilization
• Better customer satisfaction metrics

---

Future Improvements

• Real-time traffic integration
• Dynamic route optimization
• Reinforcement learning for dispatch
• Live congestion prediction
• Multi-objective network optimization

---

Tech Stack

Category	Tools
Language	Python
Data Analysis	Pandas, NumPy
Visualization	Matplotlib, Seaborn
Graph Analytics	NetworkX
ML Models	LightGBM, CatBoost, XGBoost
Graph Embeddings	Node2Vec
Notebook Environment	Jupyter

---

Conclusion

This project demonstrates how graph analytics and machine learning can transform logistics operations into a data-driven intelligent network.

The combined use of:

• Graph centrality analysis
• Corridor bottleneck detection
• ETA prediction
• Node embeddings
• Strategic FTL vs Carting optimization

creates a scalable framework for logistics network intelligence and operational decision-making.