Assessing the vulnerability of groundwater in coastal aquifers is crucial for mitigating risks associated with seawater intrusion and anthropogenic impacts. This study introduces an innovative machine learning (ML)-enhanced methodology that synergizes the strengths of two established vulnerability assessment frameworks, DRASTIC and GALDIT. The hybrid approach overcomes the limitations of each framework—DRASTIC’s inadequacies in coastal settings and GALDIT’s limited consideration of agricultural and industrial impacts. Utilizing advanced decision tree-based ML algorithms—Adaptive Boosting (AdaBoost), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LGBM), and Random Forest (RF)—this research was conducted in the Azarshahr Plain, NW Iran. Model efficacy was validated using Pearson’s correlation coefficient (r) and distance correlation (DC), with nitrate (NO3–) and total dissolved solids (TDS) serving as proxies for evaluating the DRASTIC and GALDIT models, respectively. The original DRASTIC indices exhibited weak correlations with NO3– (r = 0.24, DC = 0.25), but ML-enhanced models, particularly AdaBoost, showed significant improvements (r = 0.78, DC = 0.79). Similar enhancements were observed with GALDIT, where correlations improved markedly with AdaBoost integration. A sophisticated second-level AdaBoost meta-ensemble was developed to integrate enhanced DRASTIC and GALDIT assessments, achieving superior correlation metrics (r = 0.80, DC = 0.84 for NO3–; r = 0.82, DC = 0.83 for TDS). These results underscore the effectiveness of an integrated ML-based approach in advancing beyond traditional vulnerability assessment methods, providing a more comprehensive, accurate, and robust evaluation of coastal aquifer vulnerability.