Comparative Analysis of Classification Algorithms for Predicting Student Examination Outcomes Based on Academic Datasets

Imam Sugiharto (Universitas Bina Sarana Informatika)
Sri Indri Wahyuni (Universitas Bina Sarana Informatika)
Septian Tunijah Faradila (Universitas Bina Sarana Informatika)
Yunita Yunita (Universitas Bina Sarana Informatika)
Muhammad Ifan Rifani (Universitas Bina Sarana Informatika)

Journal of Computer Science and Informatics Engineering
Vol 5 No 3 (2026): July

Current educational data utilization remains largely focused on final evaluations, resulting in delayed interventions for students at risk of academic failure. This study compares the performance of the Support Vector Machine (SVM), K-Nearest Neighbors (K-NN), and Random Forest algorithms in classifying student examination outcomes as the foundation of an Early Warning System. A total of 6,607 Student Exam Performance records were extracted and processed using Mode Imputation, Label Encoding, and StandardScaler. The experimental results indicate that the SVM model with an RBF kernel achieved the highest accuracy of 76.40%, outperforming Random Forest (73.22%) and K-NN (71.94%). This superior performance is primarily attributed to SVM's ability to construct precise decision boundaries in high-dimensional feature spaces. Therefore, the SVM model is recommended as the primary analytical engine for the early detection of potential academic failure in educational institutions