An Optimization-Driven Artificial Neural Network Framework with Reinforcement Learning for Intelligent Phishing Email Detection
Usman Yahaya1*, 2, & 3
U Iliyasu
Sanusi Abdul Sule
Abstract
Phishing remains one of the most persistent and rapidly evolving cyber threats, requiring detection systems that are not only accurate but also adaptive to shifting attack strategies. This study proposes a hybrid phishing email detection framework that integrates an Optimization-Driven Artificial Neural Network (ANN) with Reinforcement Learning (RL) to enhance model adaptability, convergence efficiency, and decision accuracy. The ANN component learns discriminative textual and structural features extracted from a benchmark phishing dataset, while Bayesian Optimization and Particle Swarm Optimization (PSO) are employed to tune hyperparameters, reduce training variance, and improve generalization. To further address concept drift and emerging phishing patterns, an RL agent is incorporated to refine classification thresholds and adjust the model’s policy through reward-based feedback. Experimental evaluation demonstrates that the hybrid ANN–RL framework achieves superior performance compared to traditional machine-learning models, recording accuracy and F1-scores above 98% across multiple test runs. The model also shows improved resilience to misclassification, reduced false-positive rates, and faster convergence during training. The findings underscore the potential of combining optimization algorithms with reinforcement-driven adaptation to create intelligent, scalable, and self-improving phishing detection systems suitable for real-world email security environments.
References