Uniaxial compressive strength is a critical property in rock mechanics, widely used in mining, drilling, tunneling, slope stability assessment, and geotechnical design. It is a fundamental indicator of the load-bearing capacity and failure behavior of rock materials and plays a central role in engineering decision-making. Direct UCS testing, however, is costly, time-consuming, and often impractical, especially for weak, weathered, or fractured rocks where obtaining intact core samples is difficult. Laboratory preparation, standardization requirements, and equipment limitations further restrict extensive testing programs. For these reasons, reliable predictive models based on easily measurable physical and mechanical properties are highly valuable. While various predictive models exist, few studies apply ensemble machine learning approaches under both dry and saturated conditions, particularly for region-specific formations. Moisture conditions significantly influence the mechanical response of carbonate rocks, yet this factor is frequently neglected in predictive modeling. This study investigates UCS prediction from the Ilam formation in southwest Iran using physical and mechanical properties collected from limestone samples tested in both dry and saturated states. Four models random forest, deep neural network (DNN), AdaBoost, and multivariate regression (MR) were evaluated using K-fold cross-validation to ensure robust and unbiased performance assessment. Model performance was assessed via the mean square error (MSE) and coefficient of determination metrics. Results show that all models provide accurate predictions, with RF and DNN performing best under dry and saturated conditions, respectively. The study introduces a novel comparative framework and dataset for UCS prediction in Ilam limestone, supporting more efficient, reliable, and data-driven geotechnical assessments.