Our project aims to predict oxygen depletion in the Gulf of Mexico by analyzing the impact of ocean variables such as water temperature, salinity, and pH on dissolved oxygen levels, which directly influence marine health, including phytoplankton populations critical for the ecosystem. Using the NCCOS Coastal Pollutants Dataset, we employed k-nearest neighbors imputation for handling missing data and leveraged autoencoder architectures to extract latent features from the ocean variables. The best-performing model, without hidden layers, achieved high predictive accuracy (MSE of 0.0077, R² of 0.9991) when combined with XGBoost. Our SHAP analysis identified water temperature, air temperature, and barometric pressure as the most influential factors in predicting oxygen levels. These insights can help target areas at risk of oxygen depletion, guiding conservation efforts to protect marine biodiversity amidst climate change.