基于多标签分类的高血压合并症预测模型

合并症预测是典型的多标签分类问题,有效利用标签之间的相关性是提高多标签分类模型精度的关键.针对该问题提出了高血压患者常见合并症的预测模型AR-MLKNN(multi-label k-nearest neighbor based on association rules),首先从不同语义空间的临床概念中构建了患者特征表示,然后通过疾病标签关联信息量化合并症并发关系,并基于样本k邻城内标签的概率分布以后验概率的方式计算样本对每个疾病标签的隶属概率.利用合并症并发关系和疾病标签隶属概率映射形成合并症风险矩阵,基于合并症风险值,根据最小化分类损失的原则动态调整分类阈值以获取最优分类结果.实验结果表明该模型可以对高血压合并症进行较为准确的预测,F1-score达到82％,相较于常规的ML-KNN(multi-label k-nearest neighbor)模型提高了8％,在临床辅助决策领域具有一定的应用价值.

A prediction model for hypertension complications

Complication prediction is a typical multi-label classification problem.Making effective use of the correlation between labels is critical to improve the accuracy of multi-label classification models. In response to this problem, a predictive model AR-MLKNN (multi-label k-nearest neighbor based on association rules) of common complications in hypertension patients was proposed. Firstly, The characteristic representation of patients was constructed from clinical concepts in different semantic spaces.Secondly, the association information of disease labels was used to quantify the concurrent relationship of complications, and the membership probability of each disease label was calculated based on the probability distribution of k-neighborhood labels of unknown samples. Finally, the complication concurrency relationship and disease label membership probability were mapped to form a complication risk matrix. Based on the complication risk value, the classification threshold is dynamically adjusted based on the principle of minimizing the classification loss to obtain the optimal classification result. The experimental results show that the model can predict hypertension complications more accurately. The F1 -score reaches 82%, which is 8% higher than the conventional ML-KNN(multi-label k-nearest neighbor) model. Therefore it may have some application value in the field of clinical -assisted decision.