An Innovative Framework for Recommending Features in Cardiotocography for Prenatal Care

Abstract

Introduction: Health Recommender Systems (HRSs) have become essential instruments in medicine, assisting healthcare practitioners with the prompt detection of diseases and delivering tailored health advice. In the realm of maternal health, namely prenatal care, there is an increasing demand for intelligent technologies that can facilitate the precise evaluation of fetal well-being.

Objectives: This work seeks to create a customized health recommender system centered on prenatal care through the analysis of Cardiotocography (CTG) data. The aim is to forecast the probability of negative fetal outcomes and determine the most significant factors impacting fetal health during gestation.

Methods: A novel multi-objective Grey Wolf Optimization (MOGWO) algorithm was used for feature selection to enhance model accuracy and interpretability. The CTG dataset was processed to identify significant parameters, and various machine learning classifiers were trained and evaluated. Particular emphasis was placed on assessing the predictive performance of Random Forest and Decision Tree classifiers. The predictive performance of Random Forest and Decision Tree classifiers was specifically evaluated.

Results: The MOGWO-based feature selection revealed that metrics including Prolonged Decelerations (PD), Abnormal Short-Term Variability (ASTV), Abnormal Long-Term Variability (ALTV), Accelerations (AC), and Mean Long-Term Variability (MLTV) are important in predicting fetal outcomes. The Random Forest classifier has the highest classification accuracy of 95.61%, followed by the Decision Tree classifier at 93.46%.

Conclusions: The study highlights the usefulness of intelligent systems such as HRSs in prenatal care, particularly in interpreting CTG data to predict fetal well-being. This study helps pregnant mothers make better clinical decisions and reduce risk by identifying crucial diagnostic variables and using robust classifiers.