The Effect of Over-sampling and Under-sampling Techniques in Medical Datasets
Mehmet Hacibeyoglu Mohammed Hussein IBRAHIM
AbstractA well balanced dataset is crucial for the performance of the data mining classification algorithms. In medical datasets, the percentage of normal labeled classes is higher than the percentage of abnormal labeled ones, which is called as class imbalance problem in data mining. If training dataset is imbalanced, the classification algorithm generally predicts the labels of the majority class instances correctly and the minority class instances incorrectly which leads to a major problem for artificial intelligence based medical diagnosis systems. To overcome this problem, many researchers proposed over-sampling and under-sampling techniques in the literature. Over-sampling techniques increase the number of minority class instances, where the randomly chosen instances from minority class is duplicated and added to the new training dataset or synthetic instances are generated from the minority class. Under-sampling techniques decrease the number of majority class, where the randomly chosen subset of majority class is combined with the minority class instances as the new training dataset. In this study, the effect of over-sampling and under-sampling techniques in medical datasets is examined. For the experimental study, several medical benchmark datasets and well-known classification algorithms are used.