COMPARISON OF BAYESIAN MARKOV CHAIN MONTE CARLO AND MACHINE LEARNING ALGORITHMS FOR STUDENT CUMULATIVE GRADE POINT AVERAGE PREDICTION WITH FEATURE ENGINEERING

Husni Hidayat Malik (Institut Teknologi Dan Kesehatan Mahardika)
Indra Surya Permana (Institut Teknologi dan Kesehatan Mahardika)
Alva Hendi Muhammad (Universitas AMIKOM Yogyakarta)
Kusnawi Kusnawi (Universitas AMIKOM Yogyakarta)

Bulletin of Network Engineer and Informatics (BUFNETS)
Vol. 4 No. 1 (2026): BUFNETS (Bulletin of Network Engineer and Informatics) April 2026

Prediksi Indeks Prestasi Kumulatif (IPK) merupakan salah satu tantangan penting dalam manajemen akademik perguruan tinggi. Penelitian ini mengusulkan penerapan algoritma Markov Chain Monte Carlo (MCMC) berbasis inferensi Bayesian untuk memprediksi IPK mahasiswa berdasarkan Indeks Prestasi Semester (IPS) 1-8, dilengkapi dengan tiga fitur rekayasa: rata-rata IPS, variabilitas IPS (standar deviasi), dan tren IPS. Keunggulan utama pendekatan Bayesian adalah kemampuannya menghasilkan kuantifikasi ketidakpastian (uncertainty quantification) berupa interval kepercayaan 95% untuk setiap prediksi, yang tidak dapat dilakukan metode machine learning konvensional. Model MCMC dibandingkan secara komprehensif dengan tujuh algoritma machine learning: Random Forest, XGBoost, Gradient Boosting, SVM, KNN, AdaBoost, dan Bagging, menggunakan dataset 543 mahasiswa dari Institut Teknologi dan Kesehatan Mahardika periode 2017-2021. Evaluasi dilakukan melalui train-test split (80:20) dan 5-fold cross-validation menggunakan metrik RMSE, MAE, MAPE, dan R². Hasil pada data testing menunjukkan MCMC Bayesian memperoleh RMSE 0.0792 dan R² 0.867, berada pada peringkat kedua setelah Gradient Boosting. Namun pada evaluasi cross-validation yang lebih robust, MCMC Bayesian unggul dengan RMSE terendah 0.0832±0.0162 dan R² tertinggi 0.848. Temuan ini menunjukkan MCMC Bayesian tidak hanya kompetitif dari sisi akurasi, tetapi juga memberikan nilai tambah unik berupa estimasi ketidakpastian prediksi yang sangat berguna untuk sistem peringatan dini akademik.   Predicting Cumulative Achievement Index (GPA) is a key challenge in higher education academic management. This study proposes the application of Markov Chain Monte Carlo (MCMC) Bayesian inference for predicting student GPA based on Semester Achievement Index (IPS) from semesters 1-8, enriched with three engineered features: IPS mean, IPS variability (standard deviation), and IPS trend. The primary advantage of the Bayesian approach is its ability to produce uncertainty quantification in the form of 95% credible intervals for each prediction, which conventional machine learning methods cannot provide. The MCMC model was comprehensively compared against seven machine learning algorithms: Random Forest, XGBoost, Gradient Boosting, SVM, KNN, AdaBoost, and Bagging, using a dataset of 543 students from Institut Teknologi dan Kesehatan Mahardika for the period 2017-2021. Evaluation was conducted through an 80:20 train-test split and 5-fold cross-validation using RMSE, MAE, MAPE, and R² metrics. Test set results show MCMC Bayesian achieved RMSE 0.0792 and R² 0.867, ranking second after Gradient Boosting. However, in the more robust cross-validation evaluation, MCMC Bayesian outperformed all competitors with the lowest RMSE of 0.0832±0.0162 and highest R² of 0.848. These findings demonstrate that MCMC Bayesian is not only competitive in accuracy but also provides the unique value of prediction uncertainty estimation, which is highly useful for early academic warning systems