Penerapan Machine Learning untuk Prediksi Penerima Beasiswa pada Universitas Muhammadiyah Pringsewu

Qonit Zirby (Universitas Islam Lamongan)
Nur Nafi’iyah (Universitas Islam Lamongan)
Agus Setia Budi (Universitas Islam Lamongan)

Jurnal Nasional Komputasi dan Teknologi Informasi
Vol 8, No 5 (2025): Oktober 2025

Abstrak − Pendidikan tinggi berperan penting dalam pengembangan sumber daya manusia yang berkualitas, sedangkan beasiswa menjadi instrumen strategis untuk membantu mahasiswa dengan keterbatasan finansial. Namun, proses seleksi penerima beasiswa masih menghadapi tantangan efisiensi dan akurasi apabila dilakukan secara manual. Penelitian ini bertujuan membandingkan kinerja algoritma k-Nearest Neighbors (k-NN) dan Random Forest (RF) dalam klasifikasi kelayakan penerima beasiswa pada data asli dan data berimbang menggunakan Synthetic Minority Over-sampling Technique (SMOTE). Dataset diperoleh dari sumber terbuka dan melalui tahap preprocessing termasuk mengisi missing value, encoding data kategori, normalisasi data, serta penyeimbangan kelas. Hasil klasifikasi menunjukkan bahwa pada data asli, k-NN mencapai akurasi 97,34% dengan precision 0,9342, recall 1,00, dan F1-score 0,9660. Setelah penerapan SMOTE, k-NN meningkat menjadi akurasi 99,07% dengan precision 0,9835, recall 1,00, dan F1-score 0,9917. Sementara itu, algoritma RF menunjukkan performa sempurna baik pada data asli maupun data berimbang, dengan akurasi, precision, recall, dan F1-score sebesar 1,00 di kedua skenario. Hasil ini membuktikan bahwa SMOTE mampu meningkatkan performa k-NN, sedangkan RF konsisten memberikan hasil maksimal pada klasifikasi beasiswa.Kata Kunci: Beasiswa; Machine Learning; k-Nearest Neighbors; Random Forest; Synthetic Minority Over-sampling Technique (SMOTE); Abstract − Higher education plays an essential role in developing high-quality human resources, while scholarships serve as a strategic instrument to support students with financial limitations. However, the scholarship selection process still faces challenges in efficiency and accuracy when carried out manually. This study aims to compare the performance of the k-Nearest Neighbors (k-NN) and Random Forest (RF) algorithms in classifying scholarship eligibility on original and balanced datasets using the Synthetic Minority Over-sampling Technique (SMOTE). The dataset was obtained from open sources and underwent preprocessing steps, including missing value imputation, categorical data encoding, data normalization, and class balancing. The classification results show that on the original dataset, k-NN achieved an accuracy of 97.34% with a precision of 0.9342, recall of 1.00, and F1-score of 0.9660. After applying SMOTE, k-NN improved to an accuracy of 99.07% with a precision of 0.9835, recall of 1.00, and F1-score of 0.9917. Meanwhile, the RF algorithm demonstrated perfect performance in both scenarios, achieving an accuracy, precision, recall, and F1-score of 1.00. These findings confirm that SMOTE enhances the performance of k-NN, while RF consistently delivers optimal results in scholarship classification.Keywords: Scholarship; Machine Learning; k-Nearest Neighbors; Random Forest; Synthetic Minority Over-sampling Technique (SMOTE);