Application of Synthetic Minority Oversampling Technique (SMOTE) for Imbalance Class on Text Data Using kNN

Abstract

Tokopedia is one of the online marketplace providers in Indonesia that facilitates internet users to buy and sell online. Tokopedia gets an average of 147.79 million website and application visitors per month. Although it has many users, of course in an application it has advantages and disadvantages. This was conveyed by users through reviews or reviews contained in the Google Play Store. In the review, it can be seen that more users who gave 5-star rating reviews than users gave 1 star rating. The Synthetic Minority Oversampling Technique or SMOTE is a popular method applied in order to deal with class imbalances. This study aims to determine the performance of the K-Nearest Neighbor algorithm in dealing with imbalance class using Synthetic Minority Oversampling Technique (SMOTE). This study uses 5000 data consisting of 3975 negative data and 1025 positive data. Of the 5000 data divided into two parts, 70% training data and 30% test data. The SMOTE-kNN method shows a better accuracy result, which is 90% compared to using only kNN with an accuracy value of 82%.