Comparison of ensemble hybrid sampling with bagging and boosting machine learning approach for imbalanced data

Training an imbalanced dataset can cause classifiers to overfit the majority class and increase possibility of information loss for minority class. Moreover, accuracy may not give a clear picture classifier’s performance. This paper utilized decision tree (DT), support vector machine (SVM), artificial neural networks (ANN), K-nearest neighbors (KNN) Naïve Bayes (NB) besides ensemble models like random forest (RF) gradient boosting (GB), which use bagging methods, three sampling approaches seven performance metrics investigate effect imbalance on water quality data. Based results, best model was without resampling almost all except balanced accuracy, sensitivity area under curve (AUC), followed by in term specificity, precision AUC. However, sensitivity, highest achieved with under-sampling dataset. Focusing each metric separately, results showed that specificity precision, it is better preprocess classifiers. Nevertheless, improvement both when using resampled