PHISHREV: Hybrid ML and Non-Monotonic Reasoning for Phishing Detection
PHISHREV, an innovative hybrid framework, integrates machine learning classifiers with non-monotonic reasoning via Answer Set Programming (ASP) to enhance the detection of phishing websites. A post-hoc reasoning component adjusts classifier predictions through formal belief revisions, leading to a modification of 5.08% of the outputs for improved consistency. This system enables the integration of new domain knowledge in O(n) time without the need for retraining. The research tackles weaknesses in statistical ML models, which often lack contextual reasoning and are susceptible to adversarial attacks.
Key facts
- PHISHREV integrates machine learning with non-monotonic reasoning using Answer Set Programming (ASP).
- The post-hoc reasoning layer modifies 5.08% of classifier outputs.
- New domain knowledge can be added in O(n) time without model retraining.
- The framework addresses lack of contextual reasoning in statistical ML models.
- The system is designed to be robust against adversarial manipulation.
- The reasoning layer uses formal belief revisions to refine predictions.
- The approach enables context-aware decision refinement.
- The paper is published on arXiv under Computer Science > Artificial Intelligence.
Entities
Institutions
- arXiv