基于三重集成采样的胎儿健康分类

Fetal Health Classification Based on Tri-Ensemble Sampling

摘要: 胎心监护(CTG)作为监测和判断胎儿宫内健康情况的一种方法，已广泛应用于临床。产时胎心监护能连续动态的观察宫缩时胎儿在宫内的变化，提高对胎儿不良预后预测值，及时为临床做出前瞻性诊断，指导产程处理，以降低围生儿并发症和病死率。为了解决数据不平衡问题，研究采用了一种基于过采样的算法称为三重集成过采样，这一算法通过对少数类样本点进行聚类分析，将它们划分为核心点、边界点和噪音点，并随后采取不同的采样策略，扩展了边界区域的密度，并解决了少数类分布不均、新生成样本的模糊类边界等问题，旨在解决常规SMOTE过采样方法的缺点。实验结果表明，XGBoost算法在胎儿健康诊断任务中表现出色，达到了96.6%的准确率。

Abstract: Cardiotocography (CTG), as a method for monitoring and assessing the intrauterine health of fetuses, has been widely utilized in clinical practice. During childbirth, continuous monitoring of fetal heart rate dynamics allows for observation of fetal changes in utero during contractions, thereby enhancing the prediction of adverse fetal outcomes. This facilitates proactive clinical diagnosis, guides labor management, and reduces the incidence of neonatal complications and mortality. To address the issue of data imbalance, the study employed a oversampling-based algorithm known as Tri-Ensemble Sampling. This algorithm conducts cluster analysis on minority class samples, categorizing them into core points, boundary points, and noise points. Subsequently, it adopts different sampling strategies to expand the density of boundary areas, resolving issues such as uneven distribution of minority classes and ambiguous boundaries of newly generated samples, aiming to overcome the shortcomings of conventional SMOTE oversampling methods. Experimental results demonstrate outstanding performance of the XGBoost algorithm in fetal health diagnosis tasks, achieving an accuracy of 96.6%.