Research And Application On Health Supervision Risk Warning Models And Punishment Evaluation Methods

Health supervision is administrative law enforcement in the health field,involving administrative licensing,supervision and inspection,administrative punishment,and administrative coercion.Recently,with the reform of FangGuanFu,strengthening interim and ex-post supervision,enhancing health inspection regulation,and improving health administrative punishment has become the main focus.Besides,the main form of health supervision and inspection is "double random,one opening"-style random sampling of management counterparts.Random sampling may lead to the problem that management counterparts with a high risk of violation are not selected for inspection,causing regulatory loopholes.It is urgent to find regulatory objects that may have illegal behaviors from many diversified management counterparts and give early warning of their risks.At the same time,health administrative punishment involves the consistency of discretion.In order to avoid discretion unbalance and to achieve the proportionality and equivalence of punishment,it is necessary to give intelligent discretion recommendations for the fines of regulatory objects and evaluate their rationality.In the supervision and inspection,the data-driven health risk early warning model can find key regulatory objects with a high risk of violation from many management counterparts,improving the efficiency of supervision and inspection of law enforcement personnel.Moreover,it prompts the incentive of violation,achieves proper supervision,and helps evaluate discretion’s rationality in administrative punishment.In administrative punishment,it not only gives intelligent discretion recommendations for the fines of regulatory objects but also can evaluate the rationality of discretion.The research problems that need to be solved include:First,how to use big data technology to integrate Internet data based on existing information system data and build a multi-source and heterogeneous risk factor index system?Second,how to design a risk early warning model according to the data characteristics and application conditions of typical business scenarios such as public places,infectious disease prevention,and disinfection products?Model prediction can provide law enforcement personnel with high-risk management counterparts.Third,how to measure the rationality of discretion exercise in the field of administrative punishment and develop an application system for health supervision risk early warning and administrative punishment intelligent auxiliary?Given the challenges of accurately identifying medium,high-risk regulatory objects of health supervision and evaluating the rationality of discretion,this paper studies the risk early warning model and punishment evaluation method for health supervision.The designed risk early warning model meets the requirements of accuracy,explicability,and popularization;a comparative evaluation-based method is designed to judge the reasonable amount of administrative penalty in the actual health administrative law enforcement.The main contributions of this paper are as follows:1.A method of constructing a health supervision risk factor index system based on multi-source and heterogeneous data is proposed.Aiming at the problems of miscellaneous data and harrowing extraction of influential risk factors in the health supervision information system,a dynamic extraction method of high-energy risk factors extracted from the high-dimensional sparse characteristic matrix is designed by integrating Internet information.2.Three risk early warning models towards high-dimensional nonlinear are proposed.Since the high dimension of risk factors and the complex nonlinear impact of factors on the probability of violation require the model to be able to mine the nonlinear information of high-dimensional factors and resist overfitting,this paper selects XGBoost as the basic model.They were comprehensively considering the technologies that can effectively improve the effect of the model,such as anti-time-variation,sample imbalance,triple data mining,and studying the applicable conditions of these technologies,The health supervision risk early warning model is designed,including the risk early warning method of public places based on anti-time-variation,the risk early warning method of infectious disease prevention and control based on AIC-Imb-XGBoost,and the risk early warning method of disinfection products based on XGB-RGCN.3.Testing the rationality of administrative punishment based on comparative evaluation are proposed.In order to solve the problem of the inconsistent judgment of reasonable administrative penalty amount in health law enforcement,an unsupervised automatic data sample category judgment and data set construction method based on statistical data analysis,clustering,and comparative evaluation is proposed.It can learn from the historical penalty records without manual annotation,complete the judgment of the rationality of the existing penalty amount after model training and predict the penalty amount of the penalty records that have not yet made a penalty decision.4.An application system for health supervision risk early warning and administrative punishment intelligent auxiliary is developed.Based on studying the multi-source data processing technology,the method of regular iteration of risk models,the blockchain-based data security sharing scheme,and the critical technologies of the design of the Microservice+Lakehouse DevOps platform.The health supervision risk early warning and administrative punishment intelligent auxiliary application system has been developed and applied in the actual health administrative law enforcement.It promotes the new mode of health supervision based on risk early warning and targeted supervision.