Komparasi Metode SMOTE dan ADASYN dalam Meningkatkan Performa Klasifikasi Herregistrasi Mahasiswa Baru

Abstract

Universities annually accept new students at the beginning of the new school year. In the acceptance of prospective students on the Seleksi Prestasi Akademik Nasional Perguruan Tinggi Keagamaan Islam Negeri (SPAN PTKIN) di State Islamic University Of Sunan Ampel Surabaya, many prospective students who do not register will have an impact on income of the State Islamic University Of Sunan Ampel Surabaya institution. If the institution can find out early on the probability of a prospective student who will resign, then the management can take action to retain the prospective student. To overcome this, data mining classification can be carried out. The methods used in this classification are decision trees and naïve bayes. The number of students who did not re register compared to reregister resulted in the data being imbalanced. Data imbalances can affect the accuracy of the classification results. The imbalance of the data used can result in an unsuitable model. The solution to handle the data imbalance is to use the SMOTE and ADASYN oversampling methods. The purpose of this study was to compare performance of the SMOTE and ADASYN methods. The results show that the SMOTE method can balance the data in a balanced way compared to ADASYN. From the test results, the SMOTE method is more suitable to use than the ADASYN method because the ROCAUC SMOTE value is higher than ADASYN.