Comparison of machine learning algorithms to classify fetal health using cardiotocogram data

Abstract

Cardiotocogram (CTG) is one of the monitoring tools to estimate the fetus health in womb. CTG mainly yields two results fetal health rate (FHR) and uterine contractions (UC). In total, there are 21 attributes in the measurement of FHR and UC on CTG. These attributes can help obstreticians to clasify whether the fetus health is normal, suspected, or pathological. This paper compares 7 algorithms to predict the fetus health: Artificial Neural Network (ANN), Long-short Term Memory (LSTM), XG Boost (XGB), Support Vector Machine (SVM), K-Nearest Neighbour (KNN), Light GBM (LGBM), and Random Forest (RF). By employing three scenarios, this paper reports the performance measurements among those algorithms. The results show that 5 out of 7 algorithms perform very well (89-99% accurate). Those five algorithms are XGB, SVM, KNN, LGBM, RF. Furthermore, only one from five algorithm that always performs well through three scenarios: LGBM.