ACO-tuned randomized decision tree for detecting cyber threats in mobile web applications

Area of Science:

• Cybersecurity and Network Security
• Machine Learning Applications
• Mobile Computing Security

Background:

• Mobile web applications are critical for essential services but face increasing cyberattack vulnerabilities.
• Conventional security measures struggle with the dynamic and resource-constrained nature of mobile environments.
• Effective identification of intrusions in mobile systems requires advanced, adaptive solutions.

Purpose of the Study:

• To propose a novel hybrid cybersecurity framework, ACO-RDTC, for enhanced vulnerability and intrusion detection in mobile web applications.
• To introduce an adaptive feature-subset refinement mechanism tailored for mobile environments.
• To improve the efficiency and accuracy of threat detection in dynamic mobile settings.

Main Methods:

• A hybrid approach combining Ant Colony Optimisation (ACO) for global optimization and Randomised Decision Tree Classifier (RDTC) for classification.
• Data preprocessing utilizing Singular Value Decomposition (SVD) for dimensionality reduction and SelectKBest for feature selection.
• ACO dynamically optimizes hyperparameters and refines feature subsets for RDTC, enhancing its performance.

Main Results:

• The ACO-RDTC model achieved high performance on the CSE-CIC-IDS 2018 dataset: 99.08% accuracy, 98.50% precision, 99.08% recall, and 98.74% F1-score.
• The model demonstrated significant reduction in false positives compared to existing methods.
• Cross-dataset testing on UNSW-NB15 confirmed the model's generalizability with 97.82% accuracy.

Conclusions:

• The proposed ACO-RDTC framework offers a robust and scalable solution for securing modern mobile web environments.
• The adaptive feature refinement and optimization significantly improve intrusion detection capabilities.
• The study highlights the effectiveness of hybrid machine learning models in addressing complex cybersecurity challenges.