The directional overcurrent relay coordination is often performed taking into account the scenario in which the system is expected to operate most of the time. However, such an approach frequently lead to solutions that are not robust enough to deal with perturbed conditions (e.g., topology changes and line losses), which can cause a cascade sequence of unnecessary load disconnections. A new approach to improve the robustness of directional overcurrent relay coordination is presented in this work. At first, relay settings are found to the expected conditions, using a matheuristic algorithm based on Differential Evolution, Linear Programming, and local search techniques. Afterward, the solution found by the algorithm is used as an input on a re-optimization framework, in which protection must remain coordinated and fast, but undesirable load shedding is also minimized in single line dropping scenarios (N-1 criterion). The proposed approach is applied to four test cases: a real meshed network, IEEE-30 bus, IEEE-118 bus, and IEEE-300 bus. These systems have up to 590 relays and 1,044 primary/backup relations.