Measure oriented cost-sensitive SVM for 3D nodule detection

Abstract

Abstract-The class imbalance issue occurs when training a computer-aided detection (CAD) system for nodules. This imbalance causes poor prediction performance for true nodules. Moreover, the misclassification costs are different between two classes and high sensitivity of true nodules is essential in the detection. In order to eliminate or reduce the false positives while keeping high sensitivity, we present an effective wrapper framework incorporating the evaluation measure of imbalanced data into the objective function of cost sensitive SVM. We improve the performance of classification by simultaneously optimizing the best pair of misclassification cost parameter, feature subset and intrinsic parameters. We evaluated the method on a 3D Lung nodule dataset, showing that the proposed method outperforms many other exiting common methods, as well as specific imbalanced data learning methods, which indicates the effectiveness of our method on the imbalanced and unequal misclassification cost data classification.

SVM-Prot 2016: A Web-Server for Machine Learning Prediction of Protein Functional Families from Sequence Irrespective of Similarity

“…Therefore, no further reduction of the training data was made. Another effective strategy for reducing the negative influence of imbalanced data is to separately optimize the pair of cost parameters of SVM models at the same time [ 52 ], particularly the cost for the errors on the positive samples compared to negative ones. In the development of SVM models, due to the very high diversity of each training dataset (containing 7613~46,223 proteins), both the separate and uniform cost parameter optimization scheme led to very high cost parameters for both positive and negative samples that achieve similar levels of prediction performance.…”