Classification of the Weighted Network Traffic Approach Using an Optimized Deep Neural Network Algorithm

In the context of increasing network traffic type complexity, traditional traffic classification methods face significant challenges due to the nature of resource constraints, which handle massive amounts of traffic with limited processing resources. The complexity of network services is also a contributing factor, as the diversity of applications leads to a failure to adapt to new services and applications. This, in turn, results in to inaccuracies in identifying traffic types. In this paper, Weighted Network Traffic (WNT) is proposed as a means of leveraging an optimized Deep Neural Network (DNN) algorithm to enhance classification accuracy and efficiency for the entire network performance. The proposed system integrates a robust preprocessing method based on feature engineering and class reduction processes applied on IP Network Traffic Flows Labeled with 75 Apps and CICIDS2017 datasets. In the proposed WNT approach, traffic is categorized based on bandwidth metrics into three weight categories: high, moderate, and low. The optimized DNN model was evaluated using three train-test splits: 60-40, 70-30 and 80-20. The best results were achieved with the IP Network Traffic Flows Labeled with 75 Apps dataset using the 60-40 split, with a classification accuracy of 99.89%, a low loss function of 0.0043, and a and a model build time of 1 hour and 11 minutes. This performance surpasses that of the CICIDS2017 dataset and other state-of-the-art methods.