Outsmarting Spam: Resilient Model for Concept Drift and Evolving Threats
George S. Yanni, H. Salah, Fahima A. Maghraby · 2024
The ever-changing nature of spam presents a considerable challenge in maintaining effective spam detection systems. As spammers continually evolve their strategies, ensuring that spam filters remain robust and accurate becomes increasingly critical. Our proposed model is designed to address this challenge by proactively detecting and adapting to the continuously shifting landscape of spam. By leveraging advanced techniques, this method ensures sustained performance despite the non-stationary nature of streaming data. Specifically, our approach integrates a Neural Network optimized by a Genetic Algorithm with an Autoencoder. This combination enhances the model’s resilience and robustness in managing concept drift, whether in detection or mitigation. Concept drift refers to the changes in data patterns over time that can undermine the effectiveness of traditional spam detection systems. The integrated model is subjected to rigorous evaluations across both synthetic and real-world datasets, achieving a notable 95.64% accuracy on a real-life benchmark dataset, demonstrating a consistent ability to outperform existing spam detection methods. This performance not only highlights the model’s superior classification accuracy in handling imbalanced streaming data but also sets a new benchmark in the field. The innovative nature of this model underscores its significant potential for application in a variety of contexts beyond spam detection, showcasing its versatility and effectiveness in adapting to evolving data patterns.