If the packets of a flow are carried on separate physical paths for forward and reverse directions, the flow is said to be asymmetrically routed. Routing asymmetry causes problems regarding flow identification, traffic shaping, policy enforcement, quota management etc. in DPI systems. Clustering and state sharing are the two existing methods to overcome routing asymmetry. State sharing works only for stateful traffic flows whereas clustering leads to large traffic overhead. A hybrid scheme, the "Hybrid Asymmetric Traffic Classifier" (HATC) has recently been proposed to merge the better aspects of clustering and state sharing. HATC can handle both stateful and stateless flows with reduced overhead compared to clustering. In this study, we propose a novel flow identification method, the "Policy Broker-centric Traffic Classifier" (PBTC), which employs policy broker devices that undertake policy and charging rules functions in the provider's network. The performance of the proposed method is compared against that of HATC under two real traffic traces and several simulated scenarios to demonstrate that PBTC outperforms HATC.