TDT4173 – Modern Machine Learning in Practice

AIS Vessel Position Prediction

This repository contains our project for TDT4173 – Modern Machine Learning in Practice at the Norwegian University of Science and Technology (NTNU), Autumn 2024.

The project focuses on predicting future vessel positions (latitude and longitude) from AIS data. The main deliverables are two standalone Jupyter notebooks that each generate an independent set of predictions using different machine learning models.

Team name: Lanternfish

---

Authors

• Einar J. Rye
• Henrik S. Grønlund
• Theodoros Xenakis

November 2024.

---

Repository Structure

The repository is organized as follows:

• project_description.pdf
  Assignment description (self-explanatory).

• project_report.pdf
  Final project report (self-explanatory).

• limePlots/
  Folder containing LIME explanation plots used for model interpretation.

• processing/
  Contains supporting code and notebooks for:

  • Exploratory Data Analysis (EDA)
  • Feature engineering
  • Model comparison and evaluation
    These files document the development process but are not required to generate final predictions.

• Short_notebook_1.ipynb
  Standalone notebook for generating predictions using Random Forest.
  This is one of the two main entry points of the project.

• Short_notebook_2.ipynb
  Standalone notebook for generating predictions using XGBoost.
  This is the second main entry point of the project.

---

Predicition Generation

The two most important files in this repository are:

Short_notebook_1.ipynb – Random Forest

Short_notebook_2.ipynb – XGBoost

The two notebooks are fully standalone and can be run independently to:

• Load and preprocess the data
• Perform feature engineering (including shifted time-series features)
• Train separate tree models for latitude and longitude
• Apply postprocessing (nearest-port correction for land predictions)
• Generate a complete prediction file

Running Short_notebook_1.ipynb the notebook produces one full set of vessel position predictions based on Random Forest, while Short_notebook_2.ipynb produces another full set based on XGBoost.

---

The AIS data is not included in this repository due to size constraints; the authors have access to the data and can provide it upon request for replication purposes.