Imbalanced Cardiotocography Multi-classification for Antenatal Fetal Monitoring Using Weighted Random Forest

Abstract

In non-stress tests (NST) of antenatal fetal monitoring, the computerized cardiotocography (CTG) interpretation plays an important role. The digital CTG data are widely available, however, the number of abnormal cases is quite lower than that of the suspicious or normal cases. This phenomenon is referred to as imbalanced multi-classification, which has brought great challenge for machine learning to assess fetuses’ health status. Therefore, in this paper we aim to establish an intelligent evaluation model using imbalanced CTG data for monitoring fetuses’ growth. After data exploration, a weighted random forest (WRF) model was established by adjusting category weights to fulfill cost-sensitive learning. The efficiency of the proposed model was tested on the antenatal CTG dataset from the UCI repository. The WRF model achieved an average area under the receiver operating characteristic curve (ROC) of 0.99. Meanwhile, the average F1 score for the WRF (97.56%) exceeded that of the existing state-of-the-art models. The experimental results showed that the proposed model was promising for intelligent evaluating antenatal fetal health status using seriously imbalanced CTG data.