Akademik Veri Yönetim Sistemi
Avrupa Bilim ve Teknoloji Dergisi, cilt.0, sa.26, ss.22-29, 2021 (Hakemli Dergi)
The probability of encountering cyber-attacks increases with the proliferation of internet usage and the increase in the number of network devices. Intrusion detection systems are used in order to prevent the damages caused by cyber-attacks. In this study, an intrusion detection implementation based on feature selection was performed by using a convolutional neural network in order to prevent cyber-attacks. CSE-CIC-IDS2018 dataset was used during the training and testing stages. Attributes of the dataset were trained on the preprocessing layer, classification layer, and two-layer convolutional neural network. The implementation performance was assessed through accuracy, precision, and recall metrics. A retraining stage was performed in order to resolve the over-learning problem of the network. Intrusion detection was performed through synthetic data generation within the dataset. SMOTE (Synthetic Minority Over Sampling Technique) was used for synthetic data generation. In the study, Brute Force, SQL Injection, Botnet, and DoS attacks were selected as the types of threat. Attack detection accuracy of the intrusion detection system was found 98.32% and the detection accuracy obtained after retraining was found 98.8%. Following the training performed with synthetic data added into the dataset, the neural network carried out a binary classification of the data. The performance rate of detection and classification of the data as a threat was determined as 98.7% for Brute Force, 98.5% for DoS, 98.9% for Botnet, and 99.1% for SQL Injection.