Penerapan synthetic minority oversampling technique (SMOTE) pada identifikasi ikan berbasis DNA Barcoding

Abstract

Data balancing is one of the methods used to overcome imbalance amount of data in the classes to be classified. Balanced data can improve the performance of classification model. The purpose of this research was to analyze the influences of SMOTE application on the model’s accuracy in classification. Feature extraction was done using k-mers. The classification process began by building the model using Support Vector Machine (SVM) with training data obtained from the Barcode of Life Database (BOLD). The model was tested using testing data from Department of Aquatic Products Technology, Faculty of Fisheries and Marine Science, Bogor Agricultural University. The average percentage of accuracy without using SMOTE was 82.67%, while the average percentage of accuracy using SMOTE was 90.63%. These results show that the application of SMOTE on bulding the model can improve the classification accuracy of fish, both at genus and species level.