This study looks at how several hyperparameter optimization (HPO) methods—Grid Search Optimization (GSO), Random Search, and Bayesian Optimization helps in improving ship fuel consumption employing Extreme Gradient Boosting. The framework combines machine learning (XGBoost) with smart optimization to improve marine operations. The scatter plots and time-series comparisons were employed to compare each optimization approach. The statistical evaluation measures like R2, Mean Squared Error (MSE), and Mean Absolute Error (MAE) were employed to measure performance. GSO has a modest level of predictive alignment, considerable dispersion, and a greater MSE, which means there is space for improvement. Random Search shows tighter clustering and improved consistency, which leads to superior performance in both the training and testing stages. Bayesian Optimization does better than both, giving the greatest R2 values and the lowest error metrics. Its predictions are also closer to the real values. The study also looks at how well each method balances exploration and exploitation in the hyperparameter space. These results are useful for ship operators and policymakers who want to cut down on fuel expenditures and pollution. The suggested method adds to the increasing area of data-driven maritime analytics and offers a flexible and effective way to use machine learning models for operational forecasting tasks.