The　assessment　of　the　strength　of　welded　Circular　Hollow　Section　(CHS)　steel　tubular　joints　is　of　paramount　importance　in　the　safe　design　of　engineered　structures.　In　order　to　address　this　core　factor,　this　paper　employs　Gaussian　Process　Regression　(GPR),　a　probabilistic　machine　learning　method　based　on　a　collected　database,　with　the　objective　of　modelling　and　predicting　the　strength　of　three　types　of　welded　CHS　steel　tubular　joints　and　effectively　quantifying　their　associated　uncertainties.　This　study　presents　two　strength　prediction　models　for　welded　CHS　steel　tubular　joints:　a　Single-Output　Gaussian　Process　Regression　(SOGPR)　model　and　a　Multi-task　Gaussian　Process　Regression　(MTGPR)　model.　The　models　are　evaluated　and　compared　with　existing　empirical　approaches,　design　guides,　non-probabilistic　machine　learning　methods,　and　Bayesian　Linear　Regression.　The　objective　is　to　demonstrate　the　accuracy　and　high　efficiency　of　the　predictions　made　by　these　models.　Furthermore,　the　MTGPR　model　employs　the　shared　information　between　the　three　types　of　joints　to　enhance　the　prediction　performance.　Subsequently,　the　SHapley　Additive　exPlanations　method　was　employed　to　examine　the　interpretability　of　the　GPR　model　in　relation　to　the　strength　uncertainty　of　welded　CHS　steel　tubular　joints.　Ultimately,　the　uncertainty　in　the　strength　of　the　three　welded　CHS　steel　tubular　joints　is　quantified　based　on　the　proposed　prediction　methodology　in　conjunction　with　Monte　Carlo　Simulation　through　the　utilisation　of　Sobol　sensitivity　analysis　and　Morris　sensitivity　analysis.　The　findings　indicate　that　the　chord　diameter　D　and　chord　length　L　of　welded　CHS　steel　tubular　joints　have　the　greatest　impact　on　the　uncertainty　of　strength.　The　aforementioned　study　facilitates　the　optimisation　of　the　design　of　actual　engineering　structures,　the　management　of　the　range　of　strength　uncertainty　and　the　enhancement　of　the　safety　and　reliability　of　engineering　structures.