We provide a dynamic convergence technique for computing load factors and comparing vehicle system termination criteria, as well as for packet-level intrusion detection. The proposed system is able to also quickly detect attacks by combining entropy-based active learning with an attention-based model. The models created are then evaluated against typical KDD data, with and without the inclusion of an attention-based active learning mechanism in the process. The results of our experiments show that the load balancing method is superior to the other available methods. In addition, the data show how combining a pooling approach with an entropy uncertainty measure may be able to improve the decision boundary.