Klasifikasi Kelas pada Data Tidak Seimbang dalam Deteksi Mikrokalsifikasi Menggunakan Smote-Enn dan Xgboost dengan Optimasi Bayesian

Abstract

Imbalanced data is a common problem in classification. Class classification on imbalanced data can be handled with two approaches, which are data level and algorithm level. The data level is used to balance the class distribution. The algorithm level is used to improve the classification algorithm. This research is to handle class classification on imbalanced data and determine the effectiveness of hyperparameter optimization. In this research, the classification analysis of microcalcification detection is carried out which has a class imbalance of 98%. The methods used in this research are SMOTE-ENN as a method at the data level approach, XGBoost as a method at the algorithm level approach, and Bayesian optimization as a hyperparameter optimization method. Classification performance evaluation is performed by comparing XGBoost using default values and XGBoost using Bayesian optimization. The results show that the SMOTE-ENN method is able to balance the class distribution of highly imbalanced data. The XGBoost method is able to form a classification model with a high accuracy value, but low enough in precision and F-measure values. The Bayesian Optimization method was able to significantly improve the performance of the classification performance, where it succeeded in increasing the accuracy, specificity, precision, and F-measure values, but decreased the recall value. Based on the results of the analysis in this research, it is found that the XGBoost method using Bayesian optimization has a better performance evaluation than the XGBoost method using the default value.