Outlier Detection On Graduation Data Of Darussalam Gontor University Using One-Class Support Vector Machine

Abstract

Outlier detection is an important field of study because it is able to detect abnormal data distribution from a set of data. In the student graduation data, there are students with high semester GPA but do not graduate on time but students with low semester GPA can graduate on time. This study aims to detect outlier values ​​in student graduation data for the 2020-2021 class. Factors (attributes) used in this study are Student Academic Support Credit Scores (AKPAM) and Social Studies from semester one to semester six. The dataset used is 1204 graduates. The outlier detection method used is One-Class Support Vector Machine (SVM). One-class SVM is a derivative of SVM method that detects outliers based on data outside the specified class. The results of outlier detection using One-Class SVM method with three types of kernels produce the following reference values: kernel 'rbf' n by 91.4%, kernel 'linear' by 90% and kernel 'poly' by 90%. After normalization using MinMaxScaler the reference value increased by 2% in each kernel. The results of testing the One-Class SVM method get an average 90.3%, thus it can be concluded that the One-Class SVM method is feasible to be used as an outlier detection method.