#PhysicsMeetsAI: How Rabia Zoubi Unites Career, Research, and Maritime High-Tech

Balancing a master’s degree with a full-time job is a marathon. Crossing that finish line with a spectacular final grade of 1.3 while simultaneously engineering an innovative AI system for the maritime industry is nothing short of exceptional.

Our colleague Rabia Zoubi, Junior Full-Stack Dev & Data Scientist at Ship Monitor by JDS, has achieved exactly that. In his master's thesis, he dove deep into a highly complex challenge: "AI-based position prediction of ships based on historical AIS data". What makes this achievement special? His research builds a perfect bridge to the technologies and data that drive us every day at Ship Monitor by JDS.

In this interview, Rabia shares how he turned an initial development setback into a major breakthrough, why pure AI isn't always the ultimate answer, and how we supported him as an employer on the final stretch.

The Challenge: Ships Don't Move in Straight Lines

Accurate ship position prediction is critical for maritime safety, collision avoidance, and optimizing port logistics. However, reality at sea is highly unpredictable: wind, waves, and currents constantly push vessels off course. On top of that, human decisions, such as a captain's sudden maneuver, slowing down, or evading obstacles add another layer of complexity.

Until now, the state of research was divided into two distinct camps:

1. Physics-Based Models: These rely on classical laws of motion. They are highly accurate in calm waters but fail completely during unexpected maneuvers and ignore human operational patterns.
2. Pure Deep Learning Approaches: These networks learn from millions of historical tracks to identify subtle, complex patterns. However, they often ignore the laws of physics, sometimes predicting movements that are physically impossible.

The Solution: The SUPER System and "Brain Fusion"

To bridge this research gap, Rabia developed the SUPER system (Spatial, Unified Physics-Enhanced Regressor). It seamlessly integrates six classical kinematic motion models (such as Constant Velocity for open seas or Bézier curves for smooth turning maneuvers) with three neural sequence experts (LSTM, GRU, and BiGRU) working in parallel.

The Breakthrough After the Setback

The path to success was anything but linear. In his first attempt (SUPER V3), Rabia combined all 9 experts using a standard, equally weighted average. The devastating result? The prediction error spiked by +30.5% compared to the pure physics baseline. Diagnosing the issue took weeks: a phenomenon known as Expert Collapse meant that the AI and physics experts were either conflicting with or completely ignoring one another.

The turning point came with a radical paradigm shift: Brain Fusion. Instead of having each expert output a final prediction to be averaged, the experts now generate raw feature vectors. A Multi-Head Cross-Attention network then fuses these features dynamically - much like the human brain synthesizes sight, sound, and touch to make a single decision. Furthermore, a Geohash-based spatial weather attention mechanism extracts meteorological data from all 8 surrounding cells to factor in changing environmental forces.

The stunning final result: A Mean Absolute Error (MAE) of just 436 meters over a 30-minute prediction window. For a modern container ship measuring around 300 meters in length, this deviation represents a mere 1.5 ship lengths.

Big Data in Practice: The Bridge to Ship Monitor by JDS

The technological foundation of Rabia's master's thesis reads exactly like the core tech stack at Ship Monitor by JDS. The data pipeline crunched over 428 Million decoded AIS messages from the Bay of Biscay. The entire architecture for data interpolation, feature extraction, and anomaly detection was built on AWS.

When AWS cloud costs skyrocketed and strict GPU quota limits restricted parallel model training, Rabia demonstrated true engineering agility by pivoting the workload to Vast.ai, a decentralized GPU marketplace, to keep the project moving forward efficiently.

Our clients directly benefit from these breakthroughs. Merging machine learning with deep maritime domain expertise elevates predictive data quality, proactive port logistics, and situational awareness to a whole new standard.

Corporate Culture at Ship Monitor by JDS: Room for Peak Performance

A stellar master’s degree only happens when the ecosystem is right. When the double burden of full-time work, massive cloud compute experiments, and thesis writing reached its peak toward the deadline, his employer stepped in.

For us, providing this level of flexibility is a given. We strive to maintain an environment where pioneering theoretical research and practical software engineering go hand in hand.

What’s Next?

With his master’s degree officially in the bag, Rabia is looking straight ahead. At Ship Monitor by JDS, he will focus even more heavily on advanced engineering topics: robust data pipelines, stream processing with Apache Flink, scalable AWS cloud architectures, and next-generation maritime data processing.

Rabia’s ultimate piece of advice for anyone considering studying while working:

The entire team at Ship Monitor by JDS congratulates Rabia on this monumental achievement! We are incredibly proud to have you on board and look forward to building the future of maritime tech together.

Are you a master’s student, researcher, or innovation professional working toward a more transparent, sustainable, and better shipping industry? Ship Monitor by JDS actively supports research and development, and we want to empower individuals like you to drive maritime innovation. If you need high-quality AIS or maritime data samples for your thesis or project, feel free to reach out to us via our data request form:

Academic Data RequestAcademic Data Request