A Data-Driven Method Using BRB With Data Reliability and Expert Knowledge for Complex Systems Modeling
提出一种数据驱动方法，利用数据可靠性和专家知识构建置信规则库模型，通过将训练数据按可靠性聚类并集成多模型结果，提升复杂系统建模精度，并在甲状腺结节诊断中验证有效性。
The reliability of data is important for complex systems modeling. However, it remains to be a great challenge mainly due to the lack of effective measurement of data reliability. On the condition that there exist the actual result and gold standard in a piece of data simultaneously, a data-driven method is proposed using the belief rule base (BRB) with data reliability and expert knowledge. In the proposed method, gold standards are used to indicate the degree to which the actual results are correct and, thus, the reliability of each piece of data is defined as the similarity between its actual result and the corresponding gold standard. According to the reliability of each piece of data, the training dataset is divided into multiple clusters. Each cluster of data is separately used to construct and optimize a BRB model. The weight and reliability of each model are then calculated using the data in the respective cluster with the consideration of the reliabilities of the data. Finally, the estimated results of the models are integrated using the evidential reasoning rule to generate the integrated results. The proposed method is applied to help radiologists diagnose thyroid nodules, in which historical examination reports are collected from a tertiary hospital located in Hefei, Anhui, China. Its superior performance is verified by experiments.