Cyber attack detection system development with intelligent methods

As cyber-attacks have become one of the biggest problems in the world, they cause serious financial damage to countries and people. Since the increase in cyber-attacks brings along cyber-crimes, the key factors in the fight against crime and criminals are the identification of cybercriminal perpetrators and understanding the attack methods. Detecting and avoiding cyber-attacks are hard tasks. However, researchers have recently been solving these problems by developing cyber security models and making predictions with artificial intelligence methods. There are many crime prediction methods in the literature but predicting cyber-crime and cyber-attack methods has been less studied. It addresses this issue to determine an attack and the perpetrator of this type of attack using real data. The data got from law enforcement units include the type of the crime, the information of the victim, the harm and attack methods. In thesis, it analyzed cyber-crime in three different models with machine learning methods and revealed the effects of the defined features on the purpose, method, and detection of the perpetrator. Machine learning, deep learning and artificial neural networks methods used in our approach and evaluated in terms of accuracy, precision, recall and F1-score. In the attack aim prediction model, the Support Vector Machine, Decision Trees and Random Forest algorithms performed the best with 77.35% accuracy. The Nearest Neighbor algorithm showed the best performance with an accuracy rate of 69.61% in the offender prediction model. In the model that predicted the attack method, the Logistic Regression algorithm was the best with an accuracy rate of 96.69%. Results show that as the education and income level of the victim increases, the probability of cyberattack decreases. It will also make it easier to detect cyberattacks and make the fight against these attacks easier and more effective.