本项目面向大数据背景下长期护理保险的精算与风险管理，结合我国偿二代监管的实际特点，以机器学习和精算风险理论为基础，从贝叶斯高维统计推断的视角，分别对长期护理保险的定价预测模型和准备金风险进行研究。在定价预测模型方面，采用高维稀疏回归模型变量筛选方法和Bootstrap-MCMC原理，探讨高维死亡率特征元变量的提升方法，医疗费用预测的多任务Lasso建模；在保证检验功效的基础上，基于错误发现率，讨论模型设定、模型选择和预测结果的多重显著性检验；利用Python语言，探索欺诈风险识别的半监督双空间分歧协同学习模型。在准备金风险分析方面，整合图记录链接模型和动态贝叶斯网络分析方法，探究定价风险内部叠加与准备金风险外部蔓延对最终偿付能力的影响。以期在保险大数据高维统计推断的理论、精算与风险管理的机器学习模型和方法上取得进展，为我国保险公司的管理及其监管部门的决策提供科学的理论基础和有效的方法支持。

This project discusses the pricing predictive modeling and loss reserving for long-term care insurance with big data from the viewpoint of Bayesian multidimensional statistical inference, based on machine learning and modern actuarial risk theory. In terms of pricing model, this project using high dimensional sparse regression variable selection method and principle of Bootstrap-MCMC to explore high dimensional characteristics of mortality meta variable boosting scheme. discusses Multitasking Lasso modeling of healthcare costs, and model selection and model predictions of multiple significance tests based on false discovery rate. Using the Python language, this project will be exploring semi-supervised double space with high fraud risk identification differences collaborative learning model. On the Reserve risk analysis, this project integration of record linkage models dynamic Bayesian network analysis and explore pricing spread risk external risk internal stack and reserve the final impact of solvency. Finally, This project will expand high-dimensional statistical inference theory of actuarial and risk management models and methods of machine learning ,and provide a scientific rationale and effective support for China's insurance companies and their regulators.