We study the formation, evolution, and collapse of dense cores by tracking structures in a magnetohydrodynamic simulation of a star-forming cloud. We identify cores using the dendrogram algorithm and utilize machine learning techniques, including Neural Gas prototype learning and Fuzzy c-means clustering to analyze the density and velocity dispersion profiles of cores together with six bulk properties. We produce a 2-d visualization using a Uniform Manifold Approximation and Projection (UMAP), which facilitates the connection between physical properties and three partially-overlapping phases: i) unbound turbulent structures (Phase I), ii) coherent cores that have low turbulence (Phase II), and iii) bound cores, many of which become protostellar (Phase III). Within Phase II, we identify a population of long-lived coherent cores that reach a quasi-equilibrium state. Most prestellar cores form in Phase II and become protostellar after evolving into Phase III. Due to the turbulent cloud environment, the initial core properties do not uniquely predict the eventual evolution, i.e. core evolution is stochastic, and cores follow no one evolutionary path. We compare our results to NH3 observations of dense cores. Known coherent cores predominantly map into Phase II, while most turbulent pressure-confined cores map to Phase I or III. We predict that a significant fraction of observed starless cores have unresolved coherent regions and that at least 20% of observed starless cores will not form stars. Measurements of core radial profiles in addition to the usual bulk properties will enable more accurate predictions of core evolution.