Prediction of Marine Pycnocline Based on Kernel Support Vector Machine and Convex Optimization Technology

Area of Science:

• Oceanography
• Marine Data Science
• Machine Learning Applications

Background:

• Explosive growth in ocean data necessitates advanced analysis techniques.
• Incomplete ocean hydrological data due to natural factors poses challenges for analysis.
• Predicting ocean hydrological data from partial datasets is a significant area in marine science.

Purpose of the Study:

• To propose a machine learning process for handling big ocean hydrological data.
• To develop an accurate model for predicting ocean pycnocline data using incomplete datasets.
• To identify key features influencing marine density thermoclines.

Main Methods:

• Utilized a machine learning approach for processing big ocean data.
• Applied kernel function and Support Vector Machine (SVM) for nonlinear learning and convex optimization.
• Employed polynomial regression, feature scaling, and a grid search algorithm with variable step size for model training and hyperparameter tuning.

Main Results:

• An accurate model was developed to predict pycnocline in unknown domains.
• The GridSearch-SVM with variable step size demonstrated high prediction accuracy, validated by confusion matrix analysis.
• A feature ranking algorithm identified the two most influential features on the marine density thermocline.

Conclusions:

• The proposed machine learning framework effectively addresses challenges of incomplete ocean hydrological data.
• The developed SVM model provides accurate pycnocline predictions, valuable for military and scientific applications.
• This research contributes to improved understanding and prediction of oceanographic phenomena.