Earthquake declustering involves separating background events within earthquake sequences, a process that is important for seismic hazard and forecasting analysis. In this study, we apply a Supervised Machine Learning (SML) declustering algorithm to Canadian seismicity across six regions varying by tectonic behaviour. Regional parameters are estimated with the Epidemic-Type Aftershock Sequence (ETAS) model and maximum likelihood methods, and are then used to generate synthetic catalogs for training and testing the SML classifier. Declustering results are verified by comparing results from the nearest-neighbour distance (NND) method and stochastic declustering. Statistical tests including the Kolmogorov–Smirnov (KS) and Brown–Zhao (BZ) tests are used to verify independence, by confirming that background events satisfy a homogeneous Poisson process. Results show a high declustering accuracy for synthetic catalogs across all regions, outperforming previous models such as the Stochastic Declustering (SD) and NND-based Zaliapin and Ben-Zion model in every region. Estimated ETAS parameters indicate a higher earthquake productivity rate in Western Canada and the Western Canada Sedimentary Basin and a lower rate in Eastern Canada, aligning with their respective estimated aftershock ratios. Consequently, declustering results highlight tectonic variability across Canada, revealing both the strengths and limitations of the SML algorithm.