A bio inspired hybrid optimization framework for efficient real time malware detection.

Area of Science:

• Cybersecurity
• Machine Learning
• Bio-inspired Optimization

Background:

• Malware attacks via malicious URLs are rapidly increasing.
• Detecting these threats requires efficient handling of high-dimensional data.

Purpose of the Study:

• To develop a hybrid bio-inspired optimization framework for effective malware URL feature selection.
• To evaluate union (HHO∪BA) and intersection (HHO∩BA) strategies for balancing accuracy and efficiency.

Main Methods:

• Combined Harris Hawks Optimization (HHO) and Bat Algorithm (BA) for feature selection.
• Employed XGBoost and Extra Trees classifiers fine-tuned with Grid Search.
• Utilized the ISCX-URL2016 dataset for experiments.

Main Results:

• The HHO∪BA approach achieved a maximum detection accuracy of 99.52%.
• The HHO∩BA method demonstrated significantly faster training and inference times.
• Both methods showed robust classification performance on the ISCX-URL2016 dataset.

Conclusions:

• A flexible framework for malware URL detection is presented, balancing accuracy and speed.
• The HHO∪BA strategy is optimal for high-security needs, while HHO∩BA suits real-time applications.
• Findings emphasize the trade-off between detection accuracy and computational efficiency in cybersecurity.