The evaluation of mechanical strength, particularly the uniaxial compressive strength (UCS) of rocks, plays a critical role in the design and performance prediction of surface and underground structures, significantly impacting project costs and safety in engineering applications. Traditional laboratory testing methods for UCS assessment are destructive, time-consuming, and expensive, while indirect methods often lack reliability due to rock heterogeneity. This study addresses these limitations by developing advanced machine learning frameworks that integrate petrographic features with conventional rock properties to predict UCS and quantify associated uncertainties. The research utilized a comprehensive dataset from sedimentary rocks collected along Iran's southern coastlines (Persian Gulf and Gulf of Oman), encompassing mechanical properties (UCS, Brazilian tensile strength, point load index, porosity, ultrasonic pulse velocity), durability indices (Los Angeles abrasion, slake durability, aggregate impact value), and detailed petrographic characteristics derived from thin-section analysis. Three complementary approaches were implemented: (1) hybrid Neural Network-Gradient Boosting regression (ANN-GBR), (2) AutoML-optimized Random Forest, and (3) Monte Carlo simulation-based uncertainty quantification. Key petrographic features including immature and mature clastic textures, the mineral composition (quartz, chert) were used as input parameters alongside alongside laboratory testing to improve the prediction of UCS.The influence of these petrographic features on the rock’s microstructure and microcrack propagation contributes to reducing model uncertainty and enhances the reliability of predictions in complex and heterogeneous rock conditions. The AutoML-optimized Random Forest model demonstrated exceptional predictive performance with R² = 0.9884, RMSE = 0.5732 MPa, and MAPE = 3.6%, significantly outperforming traditional empirical methods. The ANN-GBR hybrid approach achieved R² = 0.9412 with RMSE = 1.385 MPa, while Monte Carlo simulations provided robust probabilistic assessments through 95% confidence intervals and systematic bias identification. Feature importance analysis revealed that soundness parameters and mineralogical composition are the most influentialpredictors, emphasizing the critical role of micro-scale petrographic properties in determining macroscopic mechanical behavior.
