ServerSecurityProject Dataset for Peer Review

Fuzzy Logic (FL) has largely been ignored in the AI revolution despite Machine Learning (ML) algorithms being energy and training inefficient. As problems become complex, ML embodies increased training times and their data centers continue to gain prominence as polluting centers. This study developed a real-world cyber security platform to detect phishing attacks, dynamically, using auto-updating feature or kernels without deploying ML to train. This study, originally, commenced with a static model platform where FL would generate an AUC of 1.0 with dynamically derived features and weights using Classification Based on Association (CBA). However, this would overfit due to old patterns and the rate of detection would eventually decline by 15% over time to an AUC of 0.85\u20130.90. The platform was updated to deploy data drift analysis to detect significant degradation (>5%) which eventually occurred due evolving phishing patterns. This led to the creation of an adaptive (or auto-updating) fuzzy system which used regenerated (or adapted) features, weights, rules and other FL aspects to ensure an AUC \u2248\u202f0.97 \u2013 1.0. This marks a departure from static ML based learning while significantly decreasing the feedback loops, training time, and enhancing detection frequency. This ensured that this model could be left unattended while the adaptive data drift did not deviate, underscoring the importance of data-driven feature tuning in maintaining detection efficacy in evolving threat landscapes.