An intelligent identification and classification system for malicious uniform resource locators (URLs)

Area of Science:

• Cybersecurity
• Machine Learning
• Data Science

Background:

• Uniform Resource Locators (URLs) are vulnerable to malicious attacks like phishing and malware.
• Detecting malicious URLs is crucial for protecting user data and system integrity.
• Existing detection methods require enhancement for improved performance.

Purpose of the Study:

• To propose a high-performance machine learning system for detecting malicious URLs.
• To develop a two-layer detection system for binary and multi-class URL classification.
• To evaluate and compare four ensemble learning approaches for malicious URL detection.

Main Methods:

• Implemented a two-layer machine learning detection system.
• Utilized four ensemble learning approaches: bagging trees (En_Bag), k-nearest neighbor (En_kNN), boosted decision trees (En_Bos), and subspace discriminator (En_Dsc).
• Evaluated models on the ISCX-URL2016 dataset using standard performance metrics.

Main Results:

• The ensemble of bagging trees (En_Bag) demonstrated superior performance rates compared to other ensemble methods.
• The ensemble of k-nearest neighbor (En_kNN) offered the highest inference speed.
• The En_Bag model achieved 99.3% accuracy in binary classification and 97.92% in multi-classification.

Conclusions:

• The proposed machine learning system effectively detects and classifies malicious URLs.
• Ensemble learning, particularly the En_Bag approach, is highly effective for malicious URL detection.
• The system offers a robust solution for identifying various types of malicious URLs, including spam, phishing, and malware.