Predicting Chronic Diseases Based on Split and Merge Method and Streaming Feature Causal Structure Learning Algorithm

摘要

Early detections of chronic diseases contribute to the prevention of such diseases. Due to the limitation of dealing with big data of BN structure learning, we figure out a new structure learning algorithm called SAM (Split And Merge) algorithm based on the thought of Adaboost, to process big data. We combine SAM algorithm with casual discovery based on the streaming features (CD-SF) algorithm to form the SAM-CD-SF algorithm. To evaluate the performance of the proposed approach, we conducted extensive experiments on the questionnaires collected from Behavior Risk Factor Surveillance System. The SAM-CD-SF can effectively deal with relative large datasets. In order to further improve the time performance, we combine SAM algorithm with casual discovery with symmetrical uncertainty based on the streaming features (CD-SU-SF) to form SAM-CD-SU-SF algorithm. Compared with SAM-CD-SF algorithm, SAM-CD-SU-SF algorithm have slightly worse accuracy, but much better time performance.