面向健康体检数据的多目标Top-k频繁模式挖掘方法  

A multi-objective Top-k frequent pattern mining approach oriented for health examination data

作　　者：邱剑锋[1,2] 武梦雨 储建军 张兴义[1,2] 苏延森 QIU Jian-feng;WU Meng-yu;CHU Jian-jun;ZHANG Xing-yi;SU Yan-sen(Key Lab of Intelligent Computing and Signal Processing of Ministry of Education,Anhui University,Hefei 230039,China;School of Computer Science and Technology,Anhui University,Hefei 230039,China;The Second People's Hospital of Hefei,Hefei 230012,China)

机构地区：[1]安徽大学计算智能与信号处理教育部重点实验室,合肥230039 [2]安徽大学计算机科学与技术学院,合肥230039 [3]合肥市第二人民医院,合肥230012

出　　处：《控制与决策》2023年第1期190-200,共11页Control and Decision

基　　金：科技部2030新一代人工智能重大项目(2018AAA0100105);国家自然科学优秀青年基金项目(61822301);国家自然科学基金项目(61822301,62076001,U1804262);安徽省自然科学基金项目(1908085MF218,2008085QF294);安徽省重点研发项目(202004j07020005);安徽高校自然科学研究项目(KJ2019A0029,KJ2021A0048,KJ2021A0634)。

摘　　要：为解决现有的模式挖掘方法没有充分利用体检数据中检查项的异常程度与特定疾病之间相关性的问题,提出一种面向健康体检数据的多目标Top-k频繁模式挖掘方法.首先,针对体检数据的特点,提出异常度和覆盖率两个指标,在此基础上,将Top-k频繁模式挖掘建模为一个多目标优化问题;其次,针对该问题,提出一种基于偏好的种群初始化策略和一个面向模式和项的双层更新策略,并基于此设计一种高效的进化多目标优化算法进行求解.实验结果表明,所提出方法所获得的Top-k个模式不仅能够有效地反映其与特定疾病之间的关联性,而且能够提供多样化的模式,为健康管理提供重要的参考依据.In order to solve the problem that the existing pattern mining methods do not make full use of the correlation between the abnormality of check items in the health examination data and specific diseases,this paper proposes a multi-objective Top-k frequent pattern mining approach oriented for health examination data.First,according to the characteristics of health examination data,two indicators of abnormality and coverage are proposed,and with these metrics,the Top-k frequent pattern mining is modeled as a multi-objective optimization problem.Then,an efficient evolutionary multi-objective optimization algorithm is designed to solve the problem,in which a preference-based population initialization strategy and a two-layer update strategy oriented to patterns and items are respectively proposed.The experimental results show that the achieved Top-k frequent patterns not only effectively reflect the correlation with the specific diseases,but also provide a variety of patterns,which gives an important reference for health management.

关 键 词：Top-k频繁模式 健康体检数据 多目标进化优化 异常度 覆盖率 基于偏好的初始化策略 双层更新策略 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]