Freight transportation networks are critical to modern supply chains but face growing challenges due to increasing trade volumes, higher service-level expectations, as well as sustainability concerns. In this context, carriers have to deal with complex planning decisions that can be assisted by solving optimization problems. This thesis focuses on the Service Network Design Problem (SNDP) and its time-dependent extension, the Scheduled Service Network Design Problem (SSNDP), NP-hard combinatorial problems that aim to jointly optimize the allocation of transportation services and shipment routing in freight transportation networks. The work addresses two major gaps in the literature: the lack of realistic and publicly available benchmark instances, and the limited scalability of existing exact solution methods when dealing with the large numbers of commodities encountered in practice. Our first contribution is an open-source instance generator capable of producing large-scale SNDP and SSNDP benchmark instances with hub-and-spoke structures. Using metrics from Complex Network Analysis, the generated networks are shown to better reflect the structural characteristics of real freight transportation systems than existing benchmarks. We then propose a new commodity aggregation-based relaxation for the SSNDP. Exploring varied aggregation strategies, we identify configurations that significantly reduce the model size of the problem while providing meaningful dual bounds. Valid inequalities to strengthen the relaxation and a repair heuristic scheme to recover primal solutions are also formulated. Computational experiments on instances containing up to 30,000 commodities demonstrate the potential of these approaches for solving large-scale SSNDP instances. Finally, our last contribution is a theoretical exact iterative disaggregation scheme based on partial commodity aggregation, together with a practical heuristic implementation. By progressively refining aggregated solutions, the practical heuristic approach enables the exploration of the tradeoff between competing considerations: model size, computational tractability, and dual bound quality. Overall, this thesis contributes new benchmark-generation tools and commodity aggregation-based optimization approaches that help extend the scale and realism of service network design instances that can be addressed.