Processing and analysis of imbalanced liver cancer patient data by case-based reasoning

摘要

The research on clinical data is one of the fastest growing fields all over the world. In general, most of the data have imbalanced issues, which may cause some problems in the researches. In this study, the methods of over-sampling and under-sampling are used for handling the issues of data imbalanced. The case based reasoning (CBR) is used for developing classification models to predict recurrent statuses of patients with liver cancer. Classification results of these two methods are compared with those of an original imbalanced dataset by the standard indicators, such as sensitivity, specificity, balanced accuracy (BAC), positive predictive value (PPV), negative predictive value (NPV), and accuracy. According to the preliminary results of classification methods, on average, the BAC of balanced methods of the under-sampling (66.07%) and the over-sampling (54.24%) exert a significant improvement compared with the imbalanced grouping dataset (48.33%). Most importantly, the under-sampling method could acquire the highest mean accuracy of the three datasets (under-sampling: 66.76%, over-sampling: 53.47%, imbalanced: 48.58%). In under-sampling method, mean PPV, NPV, and accuracy are higher than 65% (PPV: 65.44%, NPV: 69.44%, accuracy: 66.76%). The balanced datasets can provide benefits for classification models and efficiently reduce biased interpretations.