Analisis Perbandingan Teknik-teknik Oversampling Dalam Peningkatan Kinerja Model Prediksi Resiko Akademik Mahasiswa

Hutagaol, Maruli Gilbert Cristopel (2026) Analisis Perbandingan Teknik-teknik Oversampling Dalam Peningkatan Kinerja Model Prediksi Resiko Akademik Mahasiswa. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221119-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Ketidakseimbangan jumlah data pada kelas merupakan permasalahan yang umum terjadi dalam memprediksi kegagalan mahasiswa pada suatu mata kuliah, dengan jumlah mahasiswa yang tidak lulus jauh lebih sedikit dibandingkan dengan mahasiswa yang lulus. Penelitian ini bertujuan untuk membandingkan efektivitas berbagai teknik oversampling, seperti Random Over-Sampling, Random Over-Sampling Examples, Synthetic Minority Oversampling Technique-Nominal, Synthetic Minority Oversampling Technique-Encoded Nominal and Continuous, K-Means Synthetic Minority Oversampling Technique, Support Vector Machine Synthetic Minority Oversampling Technique, dan Global and Local Weighting on Synthetic Minority Oversampling Technique with Discrete features, dalam meningkatkan performa model prediksi random forest. Dataset yang digunakan terdiri atas data catatan akademik mahasiswa dari tiga program studi sarjana di sebuah perguruan tinggi swasta di Surabaya, dengan distribusi yang tidak seimbang antara kelas mayoritas dan minoritas. Evaluasi dilakukan menggunakan metrik, seperti akurasi, precision, recall, f1-score, dan area under the precison-recall curve, dan biaya untuk menilai performa model prediksi baik sebelum maupun setelah penerapan teknik oversampling. Persentase peningkatan performa prediksi dihitung dengan menghitung selisih antara nilai performa prediksi sebelum penerapan teknik oversampling dan setelah penerapan teknik oversampling, lalu membagi selisih tersebut dengan nilai performa prediksi sebelum penerapan teknik oversampling. Pada metrik recall, persentase peningkatan performa prediksi dapat mencapai nilai tertinggi, yaitu 109% dari performa prediksi sebelum penerapan teknik oversampling dengan menerapkan Global and Local Weighting on Synthetic Minority Oversampling Technique with Discrete features. Sementara itu, persentase peningkatan performa prediksi pada metrik f1-score dapat mencapai nilai tertinggi, yaitu 30% dari performa prediksi sebelum penerapan teknik oversampling dengan menerapkan Random Over-Sampling. Selain itu, persentase peningkatan performa prediksi pada metrik biaya dapat mencapai nilai tertinggi, yaitu 49% dari performa prediksi sebelum penerapan teknik oversampling dengan menerapkan Synthetic Minority Oversampling Technique-Nominal, K-Means Synthetic Minority Oversampling Technique, dan Support Vector Machine Synthetic Minority Oversampling Technique. Hasil penelitian menunjukkan bahwa penerapan teknik oversampling secara umum dapat meningkatkan performa model prediksi dalam memprediksi risiko akademik mahasiswa.

==================================================================================================================================

Data imbalance in classes is a common problem when predicting student failure in a course, as the number of students who fail is far fewer than the number of students who pass. This study aims to compare the effectiveness of various oversampling techniques, such as Random Over-Sampling, Random Over-Sampling Examples, Synthetic Minority Oversampling Technique-Nominal, Synthetic Minority Oversampling Technique-Encoded Nominal and Continuous, K-Means Synthetic Minority Oversampling Technique, Support Vector Machine Synthetic Minority Oversampling Technique, and Global and Local Weighting on Synthetic Minority Oversampling Technique with Discrete features, in improving the predictive performance of the random forest prediction model. The dataset used consists of students’ academic records from three undergraduate programs at a private university in Surabaya, with an imbalanced distribution between the majority and minority classes. The evaluation was conducted using the metrics of accuracy, precision, recall, f1-score, and area under the curve to assess the predictive performance of the prediction model both before and after applying the oversampling technique. The percentage increase in prediction performance is calculated by determining the difference between the prediction performance values before and after applying the oversampling technique and then dividing that difference by the prediction performance value before applying the oversampling technique. For the recall metric, the percentage increase in prediction performance can reach its highest value of 109%, compared to the predictive performance before applying the oversampling technique, achieved by applying Global and Local Weighting on Synthetic Minority Oversampling Technique with Discrete Features. Meanwhile, the percentage increase in predictive performance on the f1-score metric can reach its highest value of 30%, compared to the predictive performance before applying the oversampling technique, achieved by applying Random Oversampling. Furthermore, the percentage increase in prediction performance on the cost metric can reach its highest value of 49%, compared to the prediction performance before applying the oversampling technique, achieved by applying Synthetic Minority Oversampling Technique-Nominal, K-Means Synthetic Minority Oversampling Technique, and Support Vector Machine Synthetic Minority Oversampling Technique. The results of the study show that the application of oversampling techniques can generally improve the performance of predictive model in predicting student academic risk.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Oversampling, Ketidakseimbangan Kelas, Prediksi Kegagalan Mahasiswa, Random Forest, Oversampling, Class Imbalance, Student Failure Prediction, Random Forest
Subjects:	Q Science > Q Science (General) T Technology > T Technology (General)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Maruli Gilbert Cristopel Hutagaol
Date Deposited:	24 Jul 2026 23:58
Last Modified:	24 Jul 2026 23:58
URI:	http://repository.its.ac.id/id/eprint/137968

Actions (login required)

View Item