A Survival Certification Model Based on Active Learning over Medical Insurance Data

In China, Survival Certification (SC) is a work carried out for the implementation of Social Insurance (SI) policies, mainly for retirees. If a retiree is dead but his family has not notified the SI institution, then the SI institution will continue to issue pensions to the retiree. This will lead to the loss of pensions. The purpose of SC is to block the “black hole” of pension loss. However, currently, SC work mainly relies on manual services, which leads to two problems. First, due to the large number of retirees, the implementation of SC usually occupies a large amount of manpower. Secondly, at present, SC work requires all retirees to cooperate with the work of local SI institutions, while some retirees have problems with inconvenient movement or distant distances. These phenomena will lead to an increase of social costs and a waste of social resources. Thus, in this paper, a SC model based on active learning is proposed, which helps staff to narrow the scope of attention. First, we extract features from medical insurance data and analyze their effectiveness. Then, we study the effects of kinds of feature selection functions and classifiers on the SC model. The experimental results show that the model can effectively predict death and can greatly reduce the range of high-risk populations.