<![CDATA[This study aims to enhance the accuracy and generalization capability of student on-time graduation prediction by integrating the K-Nearest Neighbor (K-NN) algorithm with Particle Swarm Optimization (PSO) for parameter tuning. Historical academic records from INSTIKI were used as the primary dataset, and a 10-fold cross-validation technique was applied to ensure robust evaluation. The PSO algorithm was employed to determine the optimal k value for K-NN, with optimization parameters set to c1 = 0.5, c2 = 0.6, inertia weight w = 0.9, swarm size = 90 particles, and 100 maximum iterations. The optimized model achieved an optimal k = 23, resulting in a validation accuracy of 77.84%, outperforming the baseline K-NN’s 72.43%. In addition, improvements were observed in precision, recall, F1-score, and AUC, with the latter increasing from 0.56 to 0.68, indicating better discrimination capability. These results demonstrate that PSO effectively mitigates overfitting and enhances model stability compared to conventional K-NN. The proposed approach offers a reliable and scalable predictive model for academic early-warning systems, enabling institutions to identify at-risk students earlier and implement targeted interventions. Future work may involve incorporating non-academic features, addressing class imbalance, and exploring ensemble learning for further performance gains.]]>
Copyrights © 2025
