Energy-aware multi-objective NSGA-II optimized SVM framework for institutional admission fraud detection in educational cybercrime systems

The increasing incidence of fraudulent university admissions, coordinated agent networks, and manipulated ranking disclosures has introduced structured cyber-risk vulnerabilities in higher education systems, where existing analytical models fail to represent institution-level deception dynamics with measurable behavioral indicators. The primary limitation observed in prior studies is the dominance of student-centric or transactional fraud modeling, which lacks integration of institutional behavioral signatures such as admission pressure patterns, multi-channel promotion overlap (observed in ~ 68.3% cases), and cross-platform fraud propagation links (43% shared promotional or agent traces). To address this, the study introduces an Energy-Aware NSGA-II Optimized SVM Framework, which formalizes fraud detection as a constrained multi-objective optimization problem combining predictive maximization with computational efficiency minimization. The key novelty lies in embedding institutional cybercrime risk representation into Pareto-optimal feature selection, where behavioral, financial, and digital indicators are jointly optimized under energy-aware constraints, enabling simultaneous reduction of computational redundancy and improvement in classification stability. A domain-specific FUAD-2018–2026 dataset was constructed using verified institutional cases and validated attributes, incorporating structured evidence from regulatory alerts, admission portals, and complaint repositories with an inter-annotator agreement of 0.87, ensuring labeling consistency across five fraud categories. Methodologically, NSGA-II reduces feature dimensionality while preserving discriminative performance, and SVM with RBF kernel performs multi-class classification under fivefold validation with hyperparameter tuning via grid search. Energy profiling using CodeCarbon and EmissionTracker enables cross-platform evaluation, capturing execution variability across cloud, laptop, and desktop systems with measured energy ranges between 0.000524 and 0.000571 kWh, supporting sustainability-aware model selection. Experimental evaluation shows stable performance with 99.11% F1-score, confirming that feature reduction does not degrade classification reliability. The framework supports early institutional fraud alerting and cyber-risk monitoring in admission ecosystems by enabling structured detection of fake ranking manipulation, agent-mediated fraud, and payment irregularities. The FUAD-2018–2026 dataset, optimized feature subsets, and energy logs are designed for reproducible experimentation and extension toward large-scale educational cybercrime intelligence systems.