This work explores the optimization of a real-valued parameter, assigned to each network node running the Distributed Exponentially-weighted Flow SpliTting (DEFT) routing protocol, in order to address changes on traffic conditions. This new proposal avoids the need to alter link weights and forwarding paths, by adjusting traffic splitting. In this context, we explore the use of Evolutionary Algorithms both in single and multi-objective optimization problems, to obtain solutions that minimize network's congestion.