Comparing K-Nearest Neighbors, Random Forest and Naïve Bayes Models to Classify Fetal Health Using Resampling Methods

To preserve the lives of pregnant women and their fetuses, Cardiotocography (CTG) data is used to check the fetus’s condition and for early intervention if necessary. As a result of the existence of imbalanced classes of fetus’s condition, the performance of some models is affected when applied to real data. Accordingly, this research aims to observe the effect of data resampling methods on models’ performance by adopting the CRISP methodology. The performance of k-Nearest Neighbors (KNN), Random Forest (RF), and Naive Bayes (NB) models are compared using resampled data with several methods with the help of the CorrelationAttributeEval feature selection method for high dimension reduction. The results stated that NB and RF models’ performance has not improved with the resampling methods, while KNN has significantly improved across all metrics. RF is considered a good classifier as it was the best, with an accuracy of 93.8%. NB was the only improved model using feature selection with imbalanced data. Also, NB has the worst performance in terms of accuracy compared to KNN and RF.