An Enhanced Puma Optimized Reinforcement Learning Model for Detection of Results Anomalies in Higher Education

Abstract

The integrity of a degree depends on the accuracy and validity of examination results, which must be carefully processed and protected. The number of students being admitted to Nigerian higher education is rising yearly, making it harder for existing legacy infrastructure and the limited workforce to handle the resulting processing abnormalities. This typically leads to a significant delay in approving student results for subsequent decision-making. Unauthorized result manipulations is a common occurrence in higher education settings, and often serve as precursors to certificate counterfeiting. Given the critical role of exam administration in educational management, appropriate technologies are needed to ensure process effectiveness. Ensuring the accuracy and integrity of educational certificates and consequently preventing certificate forgery, requires that anomaly detection phase be built into result processing systems. Therefore, blockchain technology was integrated with an enhanced Puma-optimized reinforcement learning algorithm, to develop a secure and intelligent system for result filtration, storage, and protection. This platform utilized an enhanced Puma-optimized reinforcement learning algorithm in a Q-learning architecture for real-time anomaly detection. The resulting architecture was further fused with the traditional security features of blockchain technology, specifically its immutable and distributed ledger, through design, training, and testing simulations conducted in MATLAB. The improved reinforcement learning agent used a quantum superposition mutation operator to enhance the optimization process to achieve high efficiency, filtering out anomalous results in real time. To avoid local optima traps, balance exploration and exploitation, and guarantee diversity in the search for optimal parameters, the operator introduced controlled randomness. Accuracy, Precision, False Positive Rate, F1-Score, Specificity, Recall, and Detection Time were used to compare the performance of the enhanced model with those of traditional reinforcement learning, standard Puma-optimized reinforcement learning, and existing state-of-the-art works. With a 0.47% false positive rate, 99.53% specificity, 98.11% precision, 97.33% recall, 99.09% accuracy, and 42.38 milliseconds computation time across 800 epochs, the model demonstrated a high level of efficiency in detecting anomalies in students’ results.