An on-lattice kinetic Monte Carlo model of vacancy aggregation in crystalline silicon is parametrized using direct regression to evolution data from nonequilibrium molecular dynamics simulations. The approach bypasses the need to manually compute an energy barrier for each possible transition and leads to an excellent, robust representation of the molecular dynamics data. We show that the resulting lattice kinetic Monte Carlo model correctly captures the behavior of the real, continuous space system by properly accounting for continuous space entropic effects, which are often neglected in lattice-based models of atomistic processes. These contributions are particularly important at the high temperatures relevant to many steps in semiconductor materials processing.