序列模式挖掘算法在高血压药物推荐中的研究

发布时间：2018-02-13 11:30

本文关键词： 支持度阈值序列模式挖掘加权序列规则推荐高血压　出处：《太原理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：在海量的医疗数据中发现潜在的有趣的及有价值的知识,已成为当前研究的热点。针对高血压的药物治疗方案进行挖掘,发现其中隐藏的有价值的知识,为临床诊疗提供科学合理的辅助和推荐,避免医疗异常行为的发生具有重要的意义。针对以上提出的问题,本文在高血压药物治疗,序列模式挖掘,推荐算法等方面做了大量深入的研究,主要完成了以下工作:(1)首先,针对传统序列模式挖掘算法对支持度阈值静态指定的问题,提出了基于支持度阈值动态选择的序列模式挖掘算法。该算法根据频繁序列数目与支持度阈值之间的关系,采用曲线拟合的统计方法,构建了支持度阈值随频繁序列数目变化的关系模型,用于确定支持度阈值的范围,在此范围内动态选择支持度阈值。用Sign Language、BIBLE、MSNBC三个数据集进行了测试,结果表明该算法是合理有效的。(2)针对基于规则的推荐算法在高血压药物的推荐中,只考虑药物之间的相关性,而不考虑患者的其它身体指标,缺乏科学性的问题,本文提出来一种针对高血压药物治疗的加权序列规则推荐算法。该算法依据高血压危险分层对患者的处方进行加权,将挖掘得到的带有权值的序列规则用于推荐。(3)结合高血压患者的电子病历数据,基于以上两个算法,对药物治疗频繁序列及推荐进行了研究。将预处理后的加权数据进行训练,得到支持度阈值和频繁序列数目的关系模型的参数。因为挖掘的结果将作为推荐的知识库,所以这里选择支持度为0.0024,挖掘得到带有权值的频繁序列。再将频繁序列转换为带有权值的序列规则,并用于推荐。通过支持度阈值的确定和基于加权序列规则的推荐,很好的提高了高血压药物推荐的科学性和合理性,下一步工作将是推荐系统的实现。
[Abstract]:The discovery of potentially interesting and valuable knowledge in vast amounts of medical data has become a hot topic of current research. It is of great significance to provide scientific and reasonable assistance and recommendation for clinical diagnosis and treatment and to avoid the occurrence of abnormal medical behavior. The following work has been done: firstly, aiming at the problem of static specifying the support threshold of the traditional sequential pattern mining algorithm, we have done a lot of research on the recommendation algorithm. This paper presents a sequential pattern mining algorithm based on dynamic selection of support threshold, which adopts the statistical method of curve fitting according to the relationship between the number of frequent sequences and the support threshold. A relational model of support threshold varying with the number of frequent sequences is constructed, which is used to determine the range of support threshold, in which the support threshold is dynamically selected. The results show that the proposed algorithm is reasonable and effective. (2) in the recommendation of hypertension drugs, the rule-based recommendation algorithm only considers the correlation between the drugs, but does not consider the other physical indicators of the patients, which is lack of scientific problem. In this paper, a weighted sequence rule recommendation algorithm for the treatment of hypertension drugs is proposed, in which the prescriptions of patients are weighted according to the risk stratification of hypertension. The weighted sequence rule is used to recommend the EMR data of hypertension patients. Based on the above two algorithms, the frequent sequence and recommendation of drug treatment are studied. The weighted data after preprocessing are trained. The parameters of the relational model of the support threshold and the number of frequent sequences are obtained, because the results of the mining will be used as the recommended knowledge base, So the support degree is 0.0024, the frequent sequence with weight is mined, and then the frequent sequence is converted into sequence rule with weighted value, which is used for recommendation. Through the determination of support threshold and the recommendation based on weighted sequence rule, It improves the science and rationality of the recommendation of hypertension drugs, the next step will be the realization of the recommendation system.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP311.13;TP391.3

【参考文献】