Barium titanate (BaTiO3) is a ferroelectric material without toxic elements, whose ferroelectric properties such as permittivity, coercive field, and spontaneous polarization are affected by the nucleation of domains of reversed polarization and the motion of domain walls. Dislocations can act as obstacles to domain‐wall migration or as active sites for domain nucleation. Thus, studies are conducted on the utilization of dislocations to improve the ferroelectric properties of BaTiO3. However, the atomistic mechanism of domain nucleation around the dislocation core is still unclear. In this article, a machine learning (ML) potential is developed to study the influence of dislocations on domain nucleation. The potential is trained using an active‐learning approach to ensure accuracy in the bulk properties of the ferroelectric and paraelectric phases, as well as in the dislocation core structures in BaTiO3. Molecular dynamics simulations using the ML potential show that the influence of dislocations on polarization reversal depends on the directional relationship between the external electric field and the dislocation. Furthermore, strong local polarizations exist surrounding the dislocation core, owing to vacancies in the core. These polarizations can act as both domain nucleation sites and obstacles for domain migration when ordered along the dislocation line.