Research On Correlation Mining Method Of Physical Examination Data Based On Multi-objective Evolutionary Optimization

With the continuous enhancement of national health awareness,the demand for health check-ups has increased substantially.The increase in the number of physical examinations and the increasing diversification of physical examination items have resulted in the accumulation of massive physical examination data in various types of physical examination institutions at all levels,and these data often contain a wealth of medical information.How to discover hidden knowledge from physical examination data and assist doctors to implement effective health management has become a hot issue in the current research on health information mining.The current research work finds the correlation between factors or factor combinations in physical examinations and diseases by designing mining algorithms.The research results show that the multi-objective evolutionary optimization algorithm is effective in mining the correlations of physical examination data.However,in these studies,the characteristics of the physical examination data have not been fully utilized.For example,in the physical examination data,the degree of abnormality of the examination items is often related to a specific disease.In addition,the degree of aggregation of the diseased population within the value range of a certain examination item usually reflects the strong correlation between the physical examination items within the specific value range and the disease.In order to design effective data mining algorithms based on the characteristics of physical examination data,in this thesis,a Top-k frequent pattern mining algorithm based on multi-objective evolutionary optimization is proposed aiming to mine the association between abnormal items and diseases.At the same time,this thesis proposes a multi-objective evolutionary optimization based association rule mining algorithm to explore the correlation between the aggregation degree of the diseased population within the range of the physical examination items and the disease.The main research works of this thesis are introduced as follows:(1)In order to mine the association between abnormal items and diseases,this thesis proposes a Top-k frequent pattern mining algorithm based on multi-objective evolutionary optimization(MOEA-FIMED).In the physical examination data,the degree of abnormality of the examination item usually has a strong correlation with the diagnosis result.Due to the lack of consideration of this correlation,the mined k frequent patterns are weak associated with the diagnosis results and the structure is relatively similar among them,which is difficult to provide diversified frequent patterns.To this end,an abnormality index for this problem is suggested for this problem and the Top-k frequent pattern mining of physical examination data is modeled as a multi-objective optimization problem.To effectively solve this multi-objective problem,an efficient multiobjective evolutionary optimization algorithm is designed,in which a preference-based population initialization strategy and a two-layer update strategy oriented to patterns and items are proposed to improve the performance of the algorithm.The experimental results show that the frequent patterns obtained by the algorithm proposed in this thesis not only have good diversity,but also verify the correlation between abnormal items and diagnosis results through the discussion in medical literature.(2)In order to explore the correlation between the degree of aggregation of the diseased population within the numerical range of the physical examination items and the disease,this thesis proposes a multi-objective evolutionary optimization based physical examination data association rule mining algorithm(MOEA-DCC).In the physical examination data,the degree of aggregation of the diseased population within the value range of a certain physical examination item can reflect the correlation between the examination item and the disease,and the association rules mined by this correlation can help researchers to find out the risk factors and assist doctors in implementing effective health interventions.To this end,a density index is suggested to describe the correlation.Based on this index,the problem of mining association rules in physical examination data is modeled as a multi-objective optimization problem.To solve this problem,a multi-objective evolutionary optimization algorithm is designed,in which a choose-based population initialization strategy and a sorting-based crossover and mutation strategy are suggested to enhance the performance of the proposed algorithm.The former is used to improve the quality of the initial population while the latter aims to further strengthen the algorithm’s search capabilities.The experimental results verify the effectiveness of the algorithm and strategies proposed in this thesis.The correlation between the physical examination items and the disease discovered is helpful to determine the potential high-risk population related to the disease.