With the rapid expansion of e-commerce, last-mile delivery in express logistics faces significant challenges, including low efficiency and high operational costs. Taking the Xiqing District of Tianjin as a case study, this research proposes a three-stage framework integrating complex network theory and machine learning. First, the Louvain algorithm is employed to achieve intelligent partitioning of delivery areas, resulting in a modularity increase to 0.789. Second, an eXtreme Gradient Boosting (XGBoost) model is utilized to predict terminal service modes, achieving an accuracy of 87.8%. Finally, a route planning model is constructed using Particle Swarm Optimization (PSO). To validate these methods, a three-day logistics system simulation was conducted via AnyLogic to evaluate the effectiveness of different delivery policies. The results demonstrate that, compared to traditional independent delivery, the joint delivery approach reduces total costs by 25.32%. Furthermore, by introducing a carbon emission accounting model, leading to an estimated 25% reduction in daily carbon emissions, achieving a win-win situation for both economic and environmental benefits.