Addressing class imbalance problem in medical diagnosis: A genetic algorithm approach

Class imbalance problem is very common in medical diagnosis. A standard classifier might be biased towards majority class and ignore the importance of the minority class. Thus, this problem influences all supervised classification algorithms making the researchers do a lot more to deal with it. In this paper, we propose a novel sampling method called Optimized Evolutionary Under Sampling (OEUS) which uses Genetic Algorithm with optimized fitness function. The objective is to evolve biased classifier and to provide good performance on minority class. The proposed algorithm is compared with other sampling methods namely oversampling, undersampling, BOTH, and SMOTE. We also evaluate our method using several medical datasets with different weight ratios to sensitivity and specificity. The results have been contrasted by using non-parametric statistical procedure. The results reveal that the proposed method outperforms the other sampling methods and achieves a good tradeoff between data reduction and data balancing eventually producing highly precise and interpretable model in classification.