Improving classification performance for an imbalanced educational dataset example using SMOTE

Abstract

With technology, a lot of data is formed in digital environments. One of the areas with intensive data is educational data sets. Byanalyzing educational data sets, students' situatiokjgjjööÖns can be predicted by foreseeing. In this way, students can be assisted byanticipating situations such as drop-out due to failure. Educational institutions can take measures to prevent such dropouts and reducestudent drop-out. Thus, financial losses of students and educational institutions can be prevented. In this study, the data of fiveseparate associate degree students who were enrolled in Amasya University Distance Education Center in 2016-2017 were used.These are associate degree programs in child development, medical documentation and secretarial, electricity, mechatronics, andinternet and network technologies. It was estimated whether the students could graduate or not at the end of the IV. Semester withlooking at their I. and II. semester course notes. These data were analyzed by k nearest neighbor (K-NN) and KStar algorithms. Someof the data were obtained from the distance education center as imbalanced data due to the low number of students. In EducationalData Mining, researchers usually overlook the balance of the distribution on a dataset. Unbalanced data can seriously affect thesuccess of classification. Synthetic minority oversampling technique (SMOTE) method was applied to these unbalanced data and howit affected the success of classification was examined. First, the raw data were analyzed with K-nearest neighbors classifier and KStarclassifier. In this study, the analysis results of these five chapters are given in tables and comparatively. In this study, it has been seenthat SMOTE oversampling method increase the classification success. In areas where unstable data such as educational data miningmay exist, higher classification accuracy can be achieved with the help of different oversampling methods.