The Effect of Resampling on Classifier Performance: an Empirical Study

An imbalanced class on a dataset is common classification problem. The effect of using datasets can cause decrease in the performance classifier. Resampling one solutions to this This study used 100 from 3 websites: UCI Machine Learning, Kaggle, and OpenML. Each will go through processing stages: resampling process, significance testing process between evaluation values combination classifier paired t-test. Random Undersampling, Oversampling, SMOTE. Naïve Bayes Classifier, Decision Tree, Neural Network. results accuracy, precision, recall, f-measure are tested t-tests determine classifier's that were not resampled those had applied resampling. t-test also find gives significant results. obtained two first result substantially affect more than technique. second combining Network Algorithm without provides based accuracy value. Combining with SMOTE technique amount f-measure.