Imbalanced Feature Selection with Random Forest for Ovarian Cancer Diagnosis and Survival Prediction
Xiaonan Fang, Cheng Liang, Huaxiang Zhang · Journal of Computational and Theoretical Nanoscience · 2016
Ovarian cancer is the most lethal cancer of female reproductive system. Although it only ranks tenth of female malignancy tumors, its death rate is the highest among the female reproductive system tumors. Therefore, there is a pressing and consistent need to better comprehend its pathogenesis. However, the early diagnosis and survival predictions of ovarian cancer patients still remains a challenging problem today. Microarray technology has been widely accepted in early cancer diagnosis and prediction of outcome. Nevertheless, the high-dimension and imbalanced class distribution always disturb the effect of classification. In this paper, we proposed a new imbalanced feature selection method based on Random Forest called IFSRF for ovarian cancer classification. Our method selects AUC as the evaluation criterion when performing feature selection, which can relieve the negative effect of imbalanced classes. We select three manually curated ovarian cancer datasets and five widely used classifiers to show the improvement after using IFSRF. Furthermore, to demonstrate the effectiveness of the proposed method, we compare IFSRF with another widely used feature selection method Relieff. Experiments results on three ovarian cancer diagnosis and survival prediction data sets show that our feature selection method can significantly improve the AUC performance of all classifiers, especially on Random Forest. Meanwhile, the overall prediction accuracy could maintain as well.