MLOps-driven prediction of ocean oxygen saturation: Design and practical implementation

Bozkir R., CİCİOĞLU M., Calhan A.

FUTURE GENERATION COMPUTER SYSTEMS-THE INTERNATIONAL JOURNAL OF ESCIENCE, cilt.182, 2026 (SCI-Expanded, Scopus)

Özet

Oxygen saturation is a crucial parameter for sustaining marine ecosystems, as it directly affects aquatic life. Fluctuations in oxygen levels, driven by atmospheric absorption, photosynthetic activity, and water displacement, play a pivotal role in ocean dynamics. Additionally, rising temperatures and chemical pollutants further exacerbate the challenges faced by marine life. In this study, we propose a machine learning-driven framework integrated with MLOps principles to predict ocean oxygen saturation based on multiple environmental parameters, including phosphate, nitrite, nitrate, and salinity concentrations, as well as water temperature and depth. The CalCOFI dataset is employed to model the underwater environment and provide realistic data payloads for the underwater wireless sensor network (UWSN) communication scenario. The proposed approach employs machine learning models, including Ensemble, Gradient Boosting, Linear Regression, Neural Network, K-Nearest Neighbors, and Support Vector Machine, with a comprehensive performance evaluation to determine the most effective model. Furthermore, the adoption of MLOps ensures the automation of the entire model life-cycle, including training, deployment, monitoring, and continuous optimization, thereby enhancing scalability and robustness. The results demonstrate that integrating MLOps-driven machine learning pipelines significantly improves prediction accuracy and operational efficiency, making it a viable solution for real-time ocean monitoring systems. This study contributes to the advancement of intelligent environmental monitoring by leveraging AI-driven methodologies within a structured MLOps framework, ultimately facilitating proactive decision-making in marine conservation efforts.