RISKCOMMAND: MARITIME LOGISTICS & RISK ASSESSMENT

Spatiotemporal Analysis of Vessel Traffic vs Incident Clusters

Live Dashboard Simulation • Methodology • Kinematics Analysis • Strategic BI

Project Ontology

RiskCommand is a specialized analytical framework designed for maritime defense and logistics optimization. It functions as a Spatiotemporal Anomaly Detection System, processing Automatic Identification System (AIS) telemetry to identify high-risk vessel behaviors and correlate them with historical incident clusters.

Unlike traditional vessel tracking, this system integrates Unsupervised Machine Learning (Isolation Forest) to flag kinematic anomalies (unusual speed/heading patterns) and utilizes Spatial Joins to map traffic density against "Black Spot" incident zones.

Dataset & Telemetry Quality

The system processes high-frequency AIS data streams collected over a specific observation window.

• Observation Period: June 1, 2025 – June 10, 2025
• AIS Data Source: NOAA Office for Coastal Management (2025)
• Incident Data Source: NOAA Incident News Raw Data

Below is the data quality audit performed during the ingestion phase.

Data Quality Audit

Metric	Count	Percentage	Status
Total Records	8,898,171	100%	✅ Raw Ingestion
Impossible Speed (>40kn)	1,191	0.013%	⚠️ Filtered Out
Missing Coordinates	0	0.00%	✅ Clean

---

Methodology

The pipeline follows a rigorous Data Science workflow from ingestion to deployment.

graph LR
    A[Raw AIS Stream] --> B(Spatial Filtering & Cleaning)
    B --> C{Feature Engineering}
    C -->|SOG & Heading| D[Isolation Forest Model]
    C -->|Lat/Lon| E[Spatial Density Scan]
    D --> F[Anomaly Score Generation]
    E --> G[Risk Heatmap]
    F & G --> H[RiskCommand Dashboard]

Loading

1. Ingestion: Parsing massive AIS CSV logs.
2. Preprocessing: Removing physics-defying artifacts (Speed > 40kn) and null coordinates.
3. Modeling: Training an Isolation Forest on kinematic features (Speed Over Ground & Heading).
4. Deployment: Serializing the model (.joblib) for real-time inference via a Dash Web App.

---

Kinematic Anomaly Detection

The core of the analysis involves distinguishing "Normal" traffic from "Anomalous" behavior based on movement physics. The table below summarizes the statistical separation achieved by the model.

Model Statistical Separation (Speed & Heading)

Group	SOG Mean (kn)	SOG Std	SOG Max	Heading Mean (°)	Heading Std	Heading Max
Anomaly (High Risk)	24.84	5.99	99.8	152.88	116.94	359
Normal (Standard)	2.21	4.39	98.5	183.63	106.16	359

Insight: Anomalies are characterized by significantly higher average speeds (approx. 10x normal traffic) and high variance, indicating erratic maneuvers or aggressive transit.

Decision Boundary Analysis	Model Evaluation Metrics
Visualizing the separation plane between normal and anomalous vectors.	Distribution of anomaly scores across the fleet.

---

Fleet Composition Analytics

An analysis of the vessel types present in the Area of Responsibility (AOR). Codes are mapped to standard AIS vessel categories.

AIS Code	Vessel Category	Count	Distribution
37	Pleasure Craft / Sailing	2,235,763	██████████
31	Towing / Tug	1,372,241	██████
52	Tug	1,274,679	█████
36	Sailing Vessel	709,678	███
30	Fishing	688,245	███
60	Passenger	557,584	██
70	Cargo	523,236	██
80	Tanker	240,574	█
90	Other / Special	168,880	▏
57	Local / Non-Standard	152,962	▏

---

Operational Interface (Python/Dash)

The RiskCommand Interface is a real-time web application built with Plotly Dash, designed for operators to monitor live traffic and receive AI-driven alerts.

Live Geospatial Intelligence	Kinematic Anomaly Detection
Real-time map showing vessel positions, traffic density, and anomaly flags.	Scatter analysis of Speed vs. Heading to validate machine learning detection.

---

Strategic Business Intelligence (Power BI)

For executive reporting and historical trend analysis, a comprehensive Power BI Dashboard connects to the processed data pipeline.

1. Fleet Command Overview Macro-level metrics on active vessels, average network speed, and total risk alerts.	2. Anomaly Deep Dive (Explainable AI) Detailed validation of why specific vessels were flagged, analyzing their speed and heading distribution over time.
3. Spatial Risk & Incident Correlation Correlation of historical incident logs with vessel traffic density to identify high-risk "Black Spots".

---

Installation & Usage

To deploy the RiskCommand system locally:

1. Clone the repository

git clone [https://github.com/RazerArdi/maritime-logistics-risk-optimization-ais](https://github.com/RazerArdi/maritime-logistics-risk-optimization-ais)
cd maritime-logistics-risk-optimization-ais

2. Install Dependencies

pip install pandas numpy scikit-learn plotly dash dash-bootstrap-components joblib

3. Run the Dashboard

python Interface/app.py

*Access the dashboard at http://127.0.0.1:8050/*

---

Directory Structure

Project_Root/
+---Data/
|   +---Images/
|   |       DashboardPage1.png
|   |       DashboardPage2.png
|   |       decision_boundary.png
|   |       fleet_analytics.png
|   |       model_eval_boxplot.png
|   |       NOAA_logo.png
|   |       PowerBIPage1.png
|   |       PowerBIPage2.png
|   |       PowerBIPage3.png
|   |
|   +---Models/
|   |       vessel_anomaly_model.joblib  <-- Pre-trained AI Brain
|   |
|   +---Table/
|           ais_aggregated.csv
|           ais_enriched_ml.csv          <-- Main Dataset for Dashboard
|           fleet_table.csv
|           incident_risk_processed.csv  <-- Incident Data
|
+---Dataset/
|       ais-2025-06-XX.csv               <-- Raw Source Files
|       incidents.csv
|
+---Interface/
|       app.py                           <-- Main Dash Application
|
+---Notebook/
|       main.ipynb                       <-- Data Processing & Training
|
+---Reports/
        reports.pbix                     <-- Power BI Source File

---

Developed by Bayu Ardiyansyah

Informatics Engineering Student | University of Muhammadiyah Malang

Focus: Defense Technology & Data Science