Özet:

Attacks on computer networks are increasing day by day and characteristics of them are changing continuously. Network attacks destroy information security by damaging computer network systems. This situation poses a great risk for individuals, companies, institutions and even governments. To prevent or minimize the damages of network attacks, Intrusion Detection Systems are used. The development of these systems, which are created according to attack characteristics, continues parallelly to increasing attack types. In this study, it is aimed to create an intrusion detection system based on machine learning principles with anomaly detection. Recursive Feature Elimination, Forward Feature Selection, Random Forest, Decision Tree, Naive Bayes, Logistic Regression and Extreme Gradient Boosting algorithms are used during the study and evaluations are made by Accuracy, Precision, Recall and F1 Score metrics. Also, Cross Validation and ROC Curve methods are used for the evaluation. CICIDS2017 data set is used to find the most effective model by comparing the results obtained from the mentioned algorithms. As the result of this study, it is determined that the Intrusion Detection System models, which are created by classifying the features obtained the methods of Forward Feature Selection and Recursive Feature Elimination with Random Forest and Extreme Gradient Boosting algorithms, are successful.

Anahtar Kelimeler: