The economical and reliable design of steel-concrete composite structures relies on accurate predictions of the resistance of headed studs transferring the longitudinal shear forces between the two materials. The existing mechanics-based or empirical design equations do not always produce accurate and safe predictions of the stud shear resistance. This study presents the evaluation of nine machine learning (ML) algorithms and the development of optimized ML models for predicting the stud resistance. The ML models were trained and tested using databases of push-out test results for studs in both normal weight and lightweight concrete. The reliability of ML model predictions was evaluated in accordance with European and US design practices. Reduction coefficients required for the ML models to satisfy the Eurocode reliability requirements for the design shear resistance were determined. Resistance factors used in US design practice were also obtained. The developed ML models were interpreted using the SHapley Additive exPlanations (SHAP) method. Predictions by the ML models were compared with those by the existing descriptive equations, which demonstrated a higher accuracy for the ML models. A web application that conveniently provides predictions of the nominal and design stud shear resistances by the developed ML models in accordance with both European and US design practices was created and deployed to the cloud.