Hospitalization Cost Prediction for Cardiovascular Disease by Effective Feature Selection

摘要

The burden of cardiovascular diseases is increasing, and the annual growth rate of hospitalization expenses for cardiovascular diseases is much higher than that of GDP. Therefore, researchers have developed a number of intelligent systems to predict hospitalization costs for cardiovascular disease. However, there are some problems with these methods, such as the performance of real world data sets and the differences between the feature selection and the actual selection of doctors. This paper proposes a method to construct a Medical Concept Knowledge Graph (MCKG) by combining open source knowledge graphs such as Wikidata and OpenKG, open source knowledge bases such as UMLS, and doctors' prior medical knowledge. A Medical Instance Knowledge Graph (MIKG) is constructed based on MCKG and the data of cardiovascular disease related medical records from the cooperative hospital. We conduct feature selection according to MIKG, draw feature alternatives, and combine with doctor-defined rules to arrive at final feature selection. We predict hospitalization costs with random forest algorithm. Experimental results show that the average error rate of our method is lower than that of the baseline algorithms.