Predicting the long-term durability of rock in the construction of breakwaters is crucial for their safe and economic operation, but remains challenging. Here, we report on the application of Machine Learning models to such prediction. We developed a database of physical and mechanical properties of rocks from 35 rubble mound breakwaters on the Caspian Sea, Oman Sea and Persian Gulf coastlines of Iran. Properties include uniaxial compressive strength, point load strength, Brazilian tensile strength, aggregate impact and aggregate crushing values, Los Angeles abrasion, porosity, ultrasonic wave velocity, density, sodium sulfate soundness and slake durability index, together with petrophysical data. These data were analysed using the four supervised machine learning (ML) models of random forest (RF), support vector (SV) machine, gradient boost (GB) and k nearest (KN) neighbor. Model performance was assessed using RMSE computed using predicted and measured values of slake durability, and R2 of the linear regression of the predicted and measured slake durability values. The results indicate that the random forest (RF) models perform best, especially for igneous rocks: for both saturated and oven dry igneous rocks the RF model produced prediction errors of under ±0.6%, and R2 was unity to five significant figures. We conclude that ML techniques are robust methods for predicting the slake durability resistance of rock material used in the construction of breakwaters.