Self-optimized fuzzy comprehensive evaluation for real-world webrobot detection

The emergence of malicious WebRobots poses a serious threat to network security. However, detecting WebRobots in real-world scenarios remains challenging due to the impact of low-quality training samples, which can significantly reduce the accuracy of machine learning and other data-driven methods. These low-quality samples often arise from labeling difficulties and high annotation costs, leading to mislabeling and degraded model performance. To address this issue, we propose a Self-Optimizing Fuzzy Comprehensive Evaluation (SO-FCE) method, which builds upon the traditional Fuzzy Comprehensive Evaluation (FCE) framework. The core innovation lies in the integration of an iterative learning strategy that dynamically adjusts and optimizes the evaluation parameters and processes, thus mitigating the adverse effects of erroneous samples on the accuracy of the decision. This study presents a case study conducted in a real campus network, demonstrating the effectiveness of SO-FCE in handling mislabeled data. Experimental results in WebRobot detection demonstrate that SO-FCE maintains high detection performance even as sample error rates increase, unlike traditional FCE and conventional machine learning approaches, which suffer substantial performance degradation. The application of SO-FCE in real-world scenarios has shown promising results, achieving accuracy rates of 90%–99% and effectively identifying previously undisclosed robots. This highlights its significant robustness and practical value.