An advanced ensemble machine learning framework using a hierarchical stacking architecture achieved a coefficient of determination of 0.896, a root mean square error of 5.734, and a mean absolute error of 4.505 in predicting the compressive strength of engineered cementitious composites, a class of fiber-reinforced construction materials designed to overcome the brittleness of conventional concrete through tensile ductility, crack resistance, and structural durability. The model integrates Random Forest, Gradient Boosting Decision Trees, and Support Vector Regression through an optimal meta-learner to resolve nonlinear material interactions, with interpretability established through game-theoretic Shapley Additive Explanations and Partial Dependence Plots. Compressive capacity remains an indispensable determinant of structural safety, seismic performance, and infrastructure service life, but predicting ultimate compressive strength is complicated by interdependencies among raw constituent proportions, including water-to-cement ratios, mineral admixture dosages, curing regimens, and fiber geometry. Traditional mechanical compression testing according to standardized protocols like ASTM C39 provides reliable empirical measurements but is resource-intensive, labor-demanding, and limited in mapping multidimensional interrelationships across broad composition spaces. Computational modeling and machine learning have emerged as capable alternatives offering greater operational efficiency and analytical precision, though previous research showed isolated algorithms such as artificial neural networks paired with meta-heuristic optimization or gene expression programming could estimate strength in high-performance and geopolymer concretes with varying success, while individual regression architectures frequently encounter constraints including overfitting and unreliable generalization.