Abstract: Desirable properties of multi-component alloys, such as corrosion, high-temperature oxidation and irradiation resistance, are highly sensitive to the formation and migration of point defects such as vacancies and interstitials. Point defect formation, diffusion and the associated energy barriers are governed by the interactions between individual and/or groups of atoms. In this work, we use Machine learning (ML) algorithms in tandem with molecular dynamics based Nudged Elastic Band (NEB) calculations to learn the composition- and configuration-dependent formation energies and migration barriers for vacancies. Specifically, we train deep neural network models using numerical representations of the local configurational environment and also implement a physics-informed approach ensuring that the detailed balance criteria are obeyed. We discuss various scenarios including i) comparison of results using relaxed as well as unrelaxed geometries, ii) multiple methods to implement the detailed balance criteria and iii) compositional complexity and iv) configurational complexity.