XSS Saldırılarını Tespit Etmede Başarıyı Artırmak için Makine Öğrenme Tabanlı Hibrit Yaklaşım.

Saved in:

Bibliographic Details
Title:	XSS Saldırılarını Tespit Etmede Başarıyı Artırmak için Makine Öğrenme Tabanlı Hibrit Yaklaşım.
Alternate Title:	Machine Learning-Based Hybrid Approach to Increase Success in Detecting XSS Attacks.
Authors:	OLCAY, İdris¹ hermes_olcay@hotmail.com, YOLAÇAN, Esra Nergis¹
Source:	Firat University Journal of Engineering Science / Fırat Üniversitesi Mühendislik Bilimleri Dergisi. 2026, Vol. 38 Issue 1, p121-138. 18p.
Subjects:	Decision trees, Firewalls (Computer security), Internet security, False alarms, Machine learning, Detection algorithms, Ensemble learning
Abstract (English):	Cross-Site Scripting (XSS) attacks are a common type of attack that threatens the security of web applications. Today, many studies are being conducted to detect and prevent XSS attacks. One of the effective ways to prevent XSS attacks at the application layer is to use a Web Application Firewall (WAF). WAFs can generate excessive False Positives (FP) by applying strict rules at high security levels to block all potential attacks. Another disadvantage of using a WAF for web security is that attackers develop new attack vectors outside of these strict rules, which results in the WAF failing to provide the expected response. Machine learning techniques can detect XSS attacks by being trained on large datasets and recognizing patterns. In this study, 14312 benign and 12923 malicious XSS payloads were sent as HTTP requests to the WAF and six different machine learning models. The hybrid approach in this study makes decisions by combining machine learning techniques (including both individual models and ensemble methods) with the WAF's traditional attack signatures. The results of the WAF, machine learning, ensemble methods, and the hybrid approach were evaluated using accuracy, precision, sensitivity, specificity, and F1 score metrics. According to the findings, the most effective machine learning algorithm was the decision tree. According to the hybrid approach, where the WAF and decision tree were applied separately to obtain the joint result, the WAF accuracy value was improved by 30%, sensitivity by 1.27%, precision by 48%, specificity by 76%, and F1 score by 25%. [ABSTRACT FROM AUTHOR]
Abstract (Turkish):	Cross-Site Scripting (XSS, Siteler Arası Betik Çalıştırma) saldırıları, web uygulamalarının güvenliğini tehdit eden yaygın bir saldırı türüdür. Günümüzde XSS saldırılarının tespit edilmesi ve engellenmesi amacıyla birçok çalışma yapılmaktadır. XSS saldırılarını uygulama katmanında engellemenin etkili yollarından birisi Web Uygulama Güvenlik Duvarı (WAF) kullanmaktır. WAF, olası saldırıların tamamını engellemek amacıyla yüksek güvenlik seviyelerinde katı kurallar uygulayarak çok fazla Yanlış Pozitif (YP) oluşturabilmektedir. Web güvenliği için WAF kullanmanın diğer dezavantajı da saldırganlar bu katı kuralların dışında yeni saldırı vektörleri geliştirmekte ve bunun sonucunda WAF beklenen tepkiyi sağlamada başarısız olmaktadır. Makine öğrenme teknikleri, büyük veri kümeleri üzerinde eğitilerek ve desenleri tanıyarak XSS saldırılarını tespit edebilir. Çalışmada 14312 zararsız, 12923 zararlı XSS yükü HTTP isteği olarak WAF’a ve altı farklı makine öğrenme modeline gönderilmiştir. Çalışmadaki hibrit yaklaşım, makine öğrenme tekniklerini (tekil modeller ve topluluk yöntemler) WAF’ın geleneksel saldırı imzalarıyla birleştirerek karar vermektedir. WAF, makine öğrenmesi, topluluk yöntemi ve hibrit yaklaşımın sonuçları; doğruluk, kesinlik, duyarlılık, özgüllük ve F1 skoru metrikleriyle değerlendirilmiştir. Bulgulara göre, en efektif sonucu veren makine öğrenme algoritması karar ağacı olmuştur. WAF ve karar ağacının ayrı ayrı uygulanarak ortak sonucun elde edildiği hibrit yaklaşıma göre WAF’ın doğruluk değeri %30 oranında, duyarlılık %1,27 oranında, kesinlik %48 oranında, özgüllük %76 oranında ve F1 skoru da %25 oranında iyileştirilmiştir. [ABSTRACT FROM AUTHOR]
	Copyright of Firat University Journal of Engineering Science / Fırat Üniversitesi Mühendislik Bilimleri Dergisi is the property of Firat University and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database:	Engineering Source

Description
Abstract:	Cross-Site Scripting (XSS) attacks are a common type of attack that threatens the security of web applications. Today, many studies are being conducted to detect and prevent XSS attacks. One of the effective ways to prevent XSS attacks at the application layer is to use a Web Application Firewall (WAF). WAFs can generate excessive False Positives (FP) by applying strict rules at high security levels to block all potential attacks. Another disadvantage of using a WAF for web security is that attackers develop new attack vectors outside of these strict rules, which results in the WAF failing to provide the expected response. Machine learning techniques can detect XSS attacks by being trained on large datasets and recognizing patterns. In this study, 14312 benign and 12923 malicious XSS payloads were sent as HTTP requests to the WAF and six different machine learning models. The hybrid approach in this study makes decisions by combining machine learning techniques (including both individual models and ensemble methods) with the WAF's traditional attack signatures. The results of the WAF, machine learning, ensemble methods, and the hybrid approach were evaluated using accuracy, precision, sensitivity, specificity, and F1 score metrics. According to the findings, the most effective machine learning algorithm was the decision tree. According to the hybrid approach, where the WAF and decision tree were applied separately to obtain the joint result, the WAF accuracy value was improved by 30%, sensitivity by 1.27%, precision by 48%, specificity by 76%, and F1 score by 25%. [ABSTRACT FROM AUTHOR]
ISSN:	13089072
DOI:	10.35234/fumbd.1740528