This research is Part II of a two-part study that evaluates the ability of image-processing and select machine-learning algorithms to detect, classify, and track midlatitude mesoscale convective systems (MCSs) in radar-reflectivity images for the conterminous United States. This paper focuses on the tracking portion of this framework. Tracking is completed through a two-step process using slice (snapshots of instantaneous MCS intensity) data generated in Part I. The first step is to perform spatiotemporal matching, which associates slices through temporally adjacent radar-reflectivity images to generate swaths, or storm tracks. When multiple slices are found to be matches, a difference-minimization procedure is used to associate the most similar slice with the existing swath. Once this step is completed, a second step combines swaths that are spatiotemporally close. Tracking performance is assessed by calculating select metrics for all available swath-building perturbations to determine the optimal approach in tracking. Frequency maps and time series generated from the swaths suggest that the spatiotemporal occurrence of these swaths is reasonable as determined from previous work. Further, these events exhibit a diurnal cycle that is distinct from that of overall convection for the conterminous United States. Last, machine-learning predictions are found to limit areas of high MCS frequency to the central and eastern Great Plains.