A hybrid of data mining and ensemble learning forecasting for recurrent ovarian cancer

Abstract

This study applied advanced machine learning techniques and combined with ensemble learning, widely considered as the most successful method to produce objective to an inferential problem of recurrent ovarian cancer. In this study, five machine learning approaches including SVM(support vector machine), C5.0, ELM(extreme learning machine), MARS(Multivariate Adaptive Regression Splines) and RF(Random Forests) were considered to find important risk factors and to predict the recurrence-proneness for ovarian cancer. We use ensemble learning to improve the defect of classification accuracy used normal machine learning: first, selecting important risk factors by ensemble learning, then, using the five machine learning approaches to analyze again. The medical records and pathology were accessible by the Chung Shan Medical University Hospital Tumor Registry. The existing literature on recurrent ovarian cancer reveals that factors include Age, Histology, Grade, Pathologic T, Pathologic N, Pathologic M, Pathologic Stage, The International Federation of Gynecology and Obstetrics (FIGO), Surgical Margins, Performance status, CA125, Operation Optimal Debulking, Chemotherapy Guideline. There are totally 987 patients in the data set. In our study, C5.0 is the superior approach in predicting recurrence of ovarian cancer. Moreover, the classification accuracy of C5.0, MARS, RF and SVM indeed increases after using ensemble learning. Particularly the classification accuracy of C5.0 obviously improves by using ensemble learning with hybrid scheme.