Advancing Intrusion Detection Systems: a Comparative Analysis of Algorithms on the Kyoto 2015 Benchmark Dataset
Suriya Prakash J, Sreenivasa Murthy, Yash Jain, L. Lakshmanan, S Kiran · 2025
Anomaly detection is essential for identifying actions that could endanger a network and lowering the likelihood of catastrophic cyberattacks in addition to detection. It is the responsibility of Intrusion Detection Systems (IDS) to monitor particular network segments for anomalous activity, such as hacking or data theft. Recently, anomaly detection-another area of artificial intelligence-has placed a lot of emphasis on using machine learning approaches for classification. This study used the Kyoto 2015 dataset and 15 machine learning techniques, with XGBoost and Decision Tree classifiers obtaining 100 % accuracy. They demonstrated that these models are suitable for real-world implementation in cybersecurity systems, regardless of the performance parameter, with precision, recall, and$\mathbf{F 1}$values ranging from 97 % to 100 %. According to the findings, the top-performing models included Random Forest, Decision Tree, and CatBoostClassifier. In light of this, the analysis of currently available publically available cyberattack datasets will enhance and contribute to the development of NIDS frameworks in order to provide improved detection models.