Back to projects

Storm Classification (ML & Deep Learning)

ML/DL models classifying sea state and storm severity using atmospheric and oceanographic data.

Machine LearningDeep LearningTime SeriesPython

Details

About the project

I approached sea state analysis as a multi-class time-series classification problem instead of regression. Analyzed long-term atmospheric and marine data from Eastern Mediterranean (Antalya) using 36-hour sliding windows. Defined 5 severity categories: Normal, Mild, Moderate, Severe, and Extreme.

Conducted a comprehensive model comparison from classical ML (SVM, Random Forest, XGBoost) to deep learning (1D CNN, PatchTST).

Highlights

Key features

  • Temporal evolution is critical for accurate classification
  • Nonlinear classifiers essential for complex atmospheric-oceanic interactions
  • Class imbalance is a major challenge due to rare extreme events
  • PatchTST achieved the best expressiveness-generalization balance

Tech Stack

Tools used

PythonPyTorchPatchTSTCNNscikit-learnMiniROCKET