This study proposes an integrated framework that combines dynamic modeling, numerical optimization, and machine learning classification to predict the optimal design parameters of Tuned Liquid Mass Dampers (TLMDs). Two primary outputs—the optimal frequency ratio and optimal damping ratio—were analyzed using six classification models: Logistic Regression, Decision Tree, Random Forest, K Nearest Neighbors (KNN), Support Vector Machine (SVM), and Naive Bayes. Two structural configurations were examined: a single-story and a five-story shear building, each equipped with rooftop TLMDs mounted on elastomeric pads. Dynamic responses were obtained for six earthquake records using time history analysis, with liquid motion modeled by the Housner model. Optimal elastomeric pad parameters for various tank configurations were determined via the Pattern Search algorithm. The results revealed that for the optimal frequency ratio in the single-story structure, KNN and Random Forest achieved the highest F1 score (~0.73), whereas in the five-story building, prediction accuracy declined and Naive Bayes performed best (~0.68). Regarding the optimal damping ratio, Naive Bayes excelled in both structures, particularly in the five-story model. Confusion matrix analysis indicated that most errors occurred in the intermediate class, primarily due to feature overlap. By significantly reducing computational time and eliminating the need for exhaustive numerical simulations, the proposed data-driven methodology supports reliable decision-making in both preliminary and detailed stages of TLMD design. Moreover, the framework is extendable to other passive vibration control devices and more complex structural systems, advancing the concept of intelligent, efficient, and precise design tools in structural engineering.