Bilge water management is one of the very important operational and environmental challenge for merchant ships, requiring strict compliance with MARPOL Annex I regulations. This paper proposes a Smart Autonomous Bilge Management System (SABIMS) integrated together a deterministic SABIMS Logic Operations Module (SLOM) with an AI-driven decision-making module (AIDDM). A synthetic dataset. SABIMS was modelled on realistic engine room scenarios and operational benchmarks. SABIMS was trained and evaluated on synthetic data using different supervised classification algorithms: Multinomial Logistic Regression (MCLR), Decision Tree (DT), Random Forest (RF), Gradient Boosting Machine (GBM), and Extreme Gradient Boosting (XGBoost). The outcome of SABIMS training shows that XGBoost augmented with SMOTE significantly outperformed all other models, with a better class-wise performance and overall accuracy. Random Forest followed closely as second best while GBM with SMOTE followed at the third place. The assessment was done using standard evaluation metrics, viz. accuracy, precision, recall, and F1-score, with a focus on class imbalance handling and good generalization. The proposed system minimizes human intervention, while boosting predictive bilge discharge decision-making to ensure MARPOL compliance in real time. These findings highlight that synthetic data when combined with advanced machine-learning techniques, can improve operational safety and environmental sustainability in shipboard operations.