Optimized Data Processing and Genetic Algorithm based Feature Selection Method to Detect URL Phishing Attacks Using Reinforcement Learning

Abstract

Phishing is one of the major, continuously evolving cyber threats. Traditional approaches include Static Blacklist Filtering and Signature- Based Detection, suffering from a high rate of false positives and very limited adaptability to new phishing methods. To that end, RL-UPD proposes Dynamic Reinforcement Learning for Phishing Detection DRLPD and enhances data processing and feature selection through genetic algorithm while constantly updating detection parameters at runtime by reinforcement learning method. DRLPD learns new phishing trends, which raises the detection precision by 0.25%, cuts down false positive rates by 0.30%, increases efficiency by 0.20%, and enhances adaptability by 0.35% compared to traditional approaches. Reinforcement learning in this work also helps refine accuracy and efficiency while strengthening the systems against emerging threats. The novel idea in this approach shows the capability of DRLPD in revolutionizing phishing detection with machine learning, therefore opening a wide door to cybersecurity solutions. As methods of phishing become advanced, developments ensure that there will be advanced methods of detection. Exciting future work from this perspective will lie in further securing systems from phishing and contributing key insights into cybersecurity.