This paper investigates the application of machine learning techniques to a selection of offshore geotechnical engineering problems which in current industry practice is typically addressed using empirically calibrated analytical models. The specific problems covered herein are the estimation of the soil’s small strain shear stiffness from CPT data, the assessment of soil resistance to driving and the assessment of required installation pressure for suction bucket foundations. Among the several machine learning algorithms available, those belonging to the gradient boosting family were selected, as they can be effectively trained with the relatively small field datasets typically available in offshore engineering. When validated against field evidence, gradient boosting returned higher or comparable accuracy with respect to the industry standard approaches, in all the applications investigated.