SKKU-SecLab/AdFlush

Employs a feature-engineered ML classifier trained on 27 optimized features extracted from 883 candidates, achieving 0.98 F1-score on 10,000 real-world websites while reducing computational overhead by 56% CPU and 80% memory compared to prior approaches. Deployable as a Chrome Extension with robustness against adversarial manipulations (F1: 0.89–0.98) and demonstrated stability over six months without retraining. Includes comprehensive evaluation framework with support for multiple serialization formats (MOJO, ONNX) and adversarial testing datasets (GAN, GNIRTS, JavaScript obfuscators).