Classification of Audio Streaming in Network Traffic Based on Machine Learning Methods

Classifying audio streaming in network traffic has become increasingly important due to its growing prevalence. Machine learning is commonly used for this classification, but generalizing these methods across different network environments remains challenging. This research evaluates and compares the generalization of two machine learning methods, “AppScanner” and “FlowPrint,” for classifying audio streaming applications in network traffic. We assess generalization through two phases: “validation” and “experiment.” In the validation phase, where the test data only slightly differed from the training data, FlowPrint outperformed AppScanner. In the experiment phase, where the test data varied significantly from the training data, both methods exhibited a loss in generalizability for detecting known classes, with performance drops of 96.6% for FlowPrint and 89.0% for AppScanner. Despite this, both methods showed strong generalizability in detecting unknown classes, with AppScanner and FlowPrint demonstrating improvements of 14.4% and 6.4%, respectively, in detecting unknown data.