Evaluating Random Forest Performance on Packet- and Flow-Level Features for Network Traffic Classification

This paper evaluates the performance of Random Forest models for network traffic classification using two different feature sets, namely packet- and flow-level features. These features are extracted from an existing Internet traffic capture and label the data using histogram-based methods. Two distinct models are trained, one on packet-level features and the other on flow-level features. The performance of the models is assessed based on accuracy, precision, and recall, and a feature importance analysis is conducted. The results show the potential of Random Forest for effective network traffic classification and provide insights into the importance of packet- and flow-level features for such tasks.