With accelerating urbanization, traffic congestion has become a major challenge for urban mobility. Accurate identification of network bottlenecks is vital for congestion mitigation and traffic optimization. However, existing approaches relying on static topological indicators and preset thresholds often produce unstable and less reliable results. To address these limitations, a data-driven framework for bottleneck identification based on CP-CCM (congestion propagation-convergent cross mapping) model was proposed. The nonlinear causal dependencies between adjacent road segments was quantified and a congestion propagation graph was constructed. An improved breadth-first search algorithm was used to derive propagation trees and calculate each segment’s total congestion cost by integrating self-congestion and propagation effects. Segments were then ranked by congestion level to identify critical bottlenecks. SUMO simulation results showed that expanding top-level bottleneck segments increased average travel speed by 6.0% and reduced average waiting time by 25.6%, validating the accuracy and practical effectiveness of the proposed framework.