Revisiting Network Traffic Analysis: Compatible network flows for ML models

The study published on November 12, 2025, emphasizes the critical role of high-quality datasets in training machine learning models for effective cyberattack detection, especially within Internet-of-Things (IoT) networks. Researchers focused on the impact of different network traffic flow exporters on model robustness and generalization. Utilizing datasets such as Bot-IoT, IoT-23, and CICIoT23, they employed the HERA tool to analyze raw network packets, generating new labeled flows that provided more relevant features for model training. The findings indicate that direct analysis and preprocessing of PCAP files yield superior features compared to conventional CSV files, enhancing the performance of models trained for intrusion detection. This research not only contributes to the field of cybersecurity but also highlights the necessity for improved feature extraction and dataset compatibility, which are essential for developing robust machine learning applications.