With the arrivals of 3G/4G mobile networks, a diverse and new range of applications will proliferate, including video-on-demand, mobile-commerce and ubiquitous computing. It is expected a sizable proportion of these traffics move along the networks. Resources in the networks will have to be divided between voice support and data support. For the data support, multiple classes of services from the new mobile applications that have different requirements have to be monitored and managed efficiently. Traditionally Quality-of-Service (QoS) resource management was done by manual estimation of resources to be allocated in traffic profiles in GSM/GPRS environment. The resource allocations parameters are adjusted only after some period of time. In this paper, we propose a QoS resource allocation model that dynamically monitors every aspect of the network environment according to a hierarchy of QoS requirements. The model can derive knowledge of the network operation, and may even pinpoint the cause, should any anomaly occurs or malfunctions in the network. This is enabled by a hierarchy of classifiers or decision-trees, built stream-mining technology. The knowledge from the classifiers is inferred by using reasoning-of-evidence theory, and it is used for subsequent resource allocation. By this way, the resources in the network will be more dynamically and accurately adjusted, and responsive to the fluctuating traffic demands. © 2011 Springer-Verlag.