Optimizing application and algorithm complexity of machine learning methods in traffic classification

Abstract

As the Internet continues to evolve, it has become crucial for Internet Service Providers to analyze and classify their network flows. This enables them to identify suspicious activities and offer personalized services. Machine learning has been extensively deployed in network traffic classification, presenting a promising but challenging avenue. One of the primary challenges in applying machine learning to network traffic classification is reducing the computational resources used in training and implementing the model. By devising lightweight algorithms, traffic flow can be classified using fewer computational resources, effectively curtailing the escalating costs associated with the growing volume and transmission rate of traffic. In this study, we compare the performance of three classic machine learning algorithms - logistic regression, support vector machine, and shallow feedforward neural network - by employing them to classify mobile countries of origin, aiming to use as few features as possible. Remarkably, by utilizing only four features from the dataset, these three algorithms achieved an accuracy rate of 89%. This underscores the potential for computational and cost efficiency in network traffic classification with optimized machine learning methods.