Using Imbalanced Learning: A Case Study in Refractive Surgery Outcome Prediction

In the refractive surgery, the surgeon and patient will evaluate the surgery outcomes. The surgeon performs the prediction with patient's biology features, surgery parameters, theoretical formulas and hypotheses. This prediction could roughly estimate the surgery outcomes. By the popularity of refractive surgery, the clinical histories are enough to implement the surgery outcomes prediction with statistical and machine learning methods, including regression, support vector machine and neural networks. However, as the imbalanced data distribution, these data-driven methods still have drawbacks, including poor accuracy, high data size request and limited interpretability in minority class. This study introduces an over-sampling approach to improve these situation in the surgery outcome prediction. The approach over-samples the minority class to achieve better performance and accuracy. Through the experiment, it is obtained much more accurate results than the imbalanced dataset. In addition, this approach solves the result interpretability issue and the small data size issue in medical cases.