Artificial neural networks for web attack detection

[N. Stevanović]

101 list

Biografija autora: str. [103]. Bibliografija: str. [92-102].

Artificial intelligence

This dissertation presents a comprehensive approach to web attack detection using artificial neural networks. The collection of diverse malicious web traffic was carried out using honeypots, enabling the creation of a robust dataset for training models to identify cyber threats. The study addresses zero-day attack detection through various machine learning techniques capable of identifying previously unknown vulnerabilities in network traffic. To reduce catastrophic forgetting in dynamic attack environments, the use of multiple incremental learning strategies has been proposed, which enable continuous model adaptation with minimal loss of previously acquired knowledge. Тhe study introduces a population-based feature selection method, which improves classification efficiency by focusing on the most relevant network features. The dissertation presents a deep learning model for phishing email detection, based on the architectures of recurrent and convolutional neural networks. Moreover, advanced feature weighting and embedding techniques are employed to enhance phishing website detection. By integrating these methods, this dissertation provides a scalable and adaptive solution for real-time detection of web-based threats, offering significant advancements in the fields of web security and machine learning.

veštačke neuronske mreže, rekurentne neuronske mreže, konvolucione neuronske mreže, odabir atributa, ponderisanje atributa, inkrementalno učenje, sajber bezbednost, detekcija veb napada, zamke, detekcija fišinga

artificial neural networks, recurrent neural networks, convolutional neural networks, feature selection, feature weighting, incremental learning, cybersecurity, web attack detection, honeypots, phishing detection

004.8(043.3)

P 176

Tekst