This study explores the application of machine learning (ML) techniques to predict the compressive strength of concrete, with particular emphasis on mixtures incorporating both commercially and locally available materials in Mongolia. Although traditional predictive techniques like multiple linear regression (MLR), ridge, and lasso regression are widely used in structural material studies, their effectiveness may decline when working with datasets that involve regional differences and complex non-linear relationships. To address this, a decision tree model was developed and benchmarked against traditional regression methods using both newly obtained experimental data and historical datasets. Of all the models evaluated, the decision tree approach showed the highest level of accuracy, with an R2 value of 0.950 and substantially lower prediction errors (RMSE, MAPE, MAE) in comparison to linear regression models. Notably, the strength development patterns associated with Mongolian materials deviated from global trends, underscoring the necessity of region-specific modeling. In particular, overestimation in low-strength ranges and underestimation in high-strength ranges were observed, which may lead to unconservative or uneconomical design decisions in practice. These findings underscore the importance of developing a region-specific database of compressive strength tests, particularly for concrete mixes incorporating fly ash and other supplementary cementitious components. Such a database would enable the development of reliable, data-driven tools to streamline mix design and ensure structural safety across diverse construction contexts.