Studying the mechanical properties of weak rocks such as anhydrite, marl, and marl limestone is one of the important challenges in civil and mining engineering designs. After evaluating the engineering geological properties of these weak rocks at large dam sites and along a highway, this study focuses on predicting rock tensile strength (RTS). Utilizing key parameters like point load index (Is(50)), compressional wave velocity (Vp), moisture, water absorption, and porosity, various predictive models were applied, including simple regression, multivariate linear regression (MVLR), adaptive neuro-fuzzy inference system (ANFIS and support vector regression (SVR). The X-ray diffraction (XRD) and thin section assessments categorized marl limestone samples from mudstone to wackestone based on texture. The texture of marl samples was fine-grain in silt sizes. At the RTS tests, central failure modes (FMs) were identified as the predominant FMs of the marl and marl limestone samples. On the other hand, central multiple FMs were observed as the main FMs of the anhydrite samples. By increasing the calcium carbonate, the porosity, water absorption, and moisture of the samples decreased. Conversely, RTS, Is(50), Vp, and durability index increased with increasing the calcium carbonate percentage. Is(50) emerged as the most influential factor in RTS estimates and density had the least impact. Comparative analysis of modeling techniques reveals that the SVR model demonstrates superior performance, evident in coefficients of determination of 1.00, performance index of 1.97, A10-index of 1.00, and impressively low root mean square error of 0.03. These results highlight the exceptional efficacy of the SVR model in accurately estimating RTS when investigating civil project sites to reduce the sampling problem dealing with weak rocks.
