MADONNA: Browser-Based MAlicious Domain Detection Through Optimized Neural Network with Feature Analysis
Abstract
The detection of malicious domains often relies on machine learning (ML), and proposals for browser-based detection of malicious domains with high throughput have been put forward in recent years. However, existing methods suffer from limited accuracy. In this paper, we present MADONNA, a novel browser-based detector for malicious domains that surpasses the current state-of-the-art in both accuracy and throughput. Our technical contributions include optimized feature selection through correlation analysis, and the incorporation of various model optimization techniques like pruning and quantization, to enhance MADONNA’s throughput while maintaining accuracy. We conducted extensive experiments and found that our optimized architecture, the Shallow Neural Network (SNN), achieved higher accuracy than standard architectures. Furthermore, we developed and evaluated MADONNA’s Google Chrome extension, which outperformed existing methods in terms of accuracy and F1-score by six points (achieving 0.94) and four points (achieving 0.92), respectively, while maintaining a higher throughput improvement of 0.87 s. Our evaluation demonstrates that MADONNA is capable of precisely detecting malicious domains, even in real-world deployments.