The transportation operators seek for control and management strategies to push the system towards the system optimum in spite of selfish drivers who make decisions in a user optimal manner. Motivated by the Braess paradox, we propose a framework for developing optimal traffic banning strategies to improve the total travel time (TTT) of a network. More specifically, our goal is to identify the links whose closure cause the Braess paradox. This problem belongs to the family of discrete network design problem (DNDP), often formulated as a bi-level mixed-integer program. While the literature offers several algorithms to solve the mathematical problem, additional efforts are required to provide a solution method that guarantees global optimality in a reasonable computation time for real-world urban networks. We propose a heuristic optimization method based on link filtering to reduce the solution space for an exact method and evaluate its effectiveness on two sub-areas of the Chicago-Sketch network. We have designed two experiments to validate our approach and to evaluate the trade-off between computation time and solution optimality. Our results show that the proposed framework converges to the same solution as the exact method in most cases. Also, under a fixed computation time budget, the exact method could analyze only a tiny fraction of all feasible solutions, while our method not only converged significantly fast but also provided the same solution as the exact method. Lastly, we show that by implementing an optimal traffic banning strategy, the system performance can be improved up to 40%.