Analyzing Classification Algorithms for Network Intrusion Detection

Anomaly detection systems can identify unknown attacks, butless precise and frequently raise false alarms. This research looks at machine learning methods to develop intrusion detection systems that may be applied to current computer networks. First, a three-step optimization method is shown to improve the quality of the detection: 1) using improved data to rebalance the dataset, 2) training various models and 3) combining the output of the top models. The models used in this approach are trained on known attacks, and hence anomaly detection is not possible. We examined the sensitivity, accuracy, false positive rate, ROC curve, and other general classification algorithms of several binary classifiers, including Naive Bayes, Linear SVM, Random Forest, and XGBoost, to address current problems. Our findings suggested that some improvements could be made to current models. In order to defend against future attacks,we plan to utilize the Artificial Neural Network (ANN) approach, a deep learning technique, for detecting attacks and retaining them in long-term memory.