基于概率分布的临床行为检测模型

摘要

为了能对无明显时序关系或时间记录不完整的单病种的临床行为进行有效地异常检测,提出了一种基于概率分布的异常检测模型.该模型采用频率直方图和参数估计等统计学方法来确定每类临床行为的概率分布,根据确定的概率分布,采用区间选取算法来确定每类临床行为的正常的取值区间,从而生成临床行为规则库.当检测待检行为是否异常时,只需查看待检行为的统计频率是否在阚值范围内即可.通过对高血压病种降压类用药进行实验,结果表明,该模型可以有效地检测异常的临床行为.%In order to detect anomaly effectively on single disease's clinical behavior with characteristics of having no special sequential relationship or time recording being not complete, an anomaly detection model based on probility distribution model is advanced. The method adoptes some statistics methods such as frequence histogram and parameter estimate, and so on, to ensure the probility distribution of clinical behavior. Adopt section choosing algorithm to ensure normal value range of each kind of clinical behavior according to the probility distribution ensured, and the rule base is obtained. When testing the behavior for inspection, we only need to check whether its statistic frequence is in the normal range or not. The experimental result on the usage of hypotensor indicates that the model can test the anomaly effectively.