Running summarizing tools on a new article

Abstract: This work applies machine learning techniques to improve network intrusion detection using the CIC-IDS-2017 dataset, which includes realistic examples of both normal and attack traffic. Traditional intrusion detection systems rely on static signatures, making it difficult to recognize new or evolving attack patterns. To overcome this limitation, the dataset was preprocessed, analyzed, and optimized using Recursive Feature Elimination (RFE) to identify the most relevant flow-based features for anomaly detection. The results show that features related to average packet size, forward and backward packet length behavior, packet timing, destination ports, and total packet volume provide the highest impact on model performance. Several supervised learning algorithms were trained and compared, including Logistic Regression, KNN, Decision Tree, Random Forest, XGBoost, CatBoost, and a Voting Classifier. The evaluation demonstrates that ensemble models consistently achieve over 99% accuracy, outperforming simpler classifiers. These findings suggest that feature-optimized machine learning approaches can significantly improve intrusion detection accuracy and support future development of real-time detection systems.