The importance of network traffic classification has grown over the last decade. Coupled with advances in software and theory, the range of classification techniques has also increased. Network operators can predict demands in future traffic to high accuracy and better identify anomalous behavior. Multiple machine learning tools have been developed in this field and each have had varying degrees of success. In this paper we use supervised machine learning within a frequentist neural network to develop a model capable of achieving high classification accuracy and maintaining low system throughput. We will compare our model to previous work on Bayesian neural networks and other standard classification techniques in the context of real-time classification. The spatial and temporal stabilities of the different models will also be compared. Finally, we investigate the relationship between the convergence times of each model and the size of training dataset. Emphasis will be placed on experimental design and methodology to adequately justify and contextualize our analysis, as well as clarify the limitations of our results. Challenges in the field and areas for further work will also be discussed.