Boosted SVM for extracting rules from imbalanced data in application to prediction of the post-operative life expectancy in the lung cancer patients

• We generalize the learning task for SVM by introducing individual penalty parameters for each example.
• We propose hybrid approach which is based on combining AdaBoost algorithm with SVM for imbalanced data phenomenon. We call this boosted SVM.
• The proposed hybrid approach minimizes weighted exponential error function.
• Once the boosted SVM obtains high predictive accuracy, it seems proper to use it as an oracle to re-label examples and use the new dataset for the rules induction in order to obtain an interpretable model.
• We present the application of the boosted SVM to the medical domain, that is, the prediction of the post-operative life expectancy in the lung cancer patients.

In this paper, we present boosted SVM dedicated to solve imbalanced data problems. Proposed solution combines the benefits of using ensemble classifiers for uneven data together with cost-sensitive support vectors machines. Further, we present oracle-based approach for extracting decision rules from the boosted SVM. In the next step we examine the quality of the proposed method by comparing the performance with other algorithms which deal with imbalanced data. Finally, boosted SVM is used for medical application of predicting post-operative life expectancy in the lung cancer patients.

Figure optionsDownload as PowerPoint slide