Adaptive Anomaly Detection Algorithm for Real-Time Identification of Phishing and Credential Stuffing Attacks in E-Learning Systems Using Deep Reinforcement Learning

The quick embrace of e-learning software programs and cloud-based LMS has greatly enhanced digital education through better access and scalability of the platform. Nevertheless, with the reliance on the use of online authentication platforms, there is also a higher susceptibility to cyberattacks such as phishing attacks and credential stuffing attacks. These attacks utilize vulnerabilities in the user authentication process, stolen credentials, and deceptive communication tactics to gain unauthorized access, steal academic credentials, and disrupt the entire learning process. Traditional methods of intrusion detection and static machine learning have proven less effective in adapting to dynamic attack patterns and have a high percentage of false positives. To mitigate these problems, this paper proposes a novel Adaptive Anomaly Detection Algorithm that uses Deep Reinforcement Learning (DRL) to Real-Time Identify Phishing and Credential Stuffing Attacks in E-Learning Systems. The proposed approach involves the use of behavioral analysis, sequential anomaly detection, and adaptive policy optimization. Data pre-processing, behavioral feature extraction, state-action-based deep reinforcement learning, attack categorization, and adaptive mechanisms will form the basis of the proposed algorithm. Among behavioral features considered are failed login attempts, rate of IP change, session time, URL reputation, and device mismatch. Evaluation through experiments was carried out using a mixed cybersecurity data set that contained 125,000 instances with phishing attacks, credential stuffing, and legitimate users' actions. The developed DRL model attained outstanding results with an accuracy of 98.4%, precision of 98.1%, recall of 97.8%, and F1-score of 97.9% while having a minimum of false positives of 1.9%. A comparative study indicates remarkable improvements compared to SVM, Random Forest, CNN, and LSTM techniques. The findings prove the effectiveness of adaptive reinforcement learning in improving cybersecurity capabilities in contemporary e-learning environments.

Hali tarjima qilinmagan