부산대학교 도서관

There are three challenges in real-life disease detection scenarios: 1) the number of open samples is small; 2) the difficulty and cost of labeling the samples are very high; 3) The class distribution of the samples is extremely unbalanced. To solve these problems, we combine the Synthetic Minority Oversampling TEchnique (SMOTE) with the Fuzziness based Random Vector Functional Link network (F-RVFL) and propose an Improved F-RVFL algorithm (IF-RVFL) in this paper. The proposed IF-RVFL is a semi-supervised learning algorithm using the self-training strategy, which can make full use of a large number of unlabeled samples to improve the performance of the model. At the same time, the SMOTE technique enables the IFRVFL to effectively solve the class imbalanced problem. The effectiveness of the proposed IF-RVFL has been verified on a reallife liver disease data set. Extensive experimental results show that the IF-RVFL algorithm can achieve better generalization ability than the RVFL, F-RVFL, and their variants. IF-RVFL also provides a new technique with great potential for other disease detection.