Classification of Fetal Health Using the K-Nearest Neighbor Method and the Relieff Feature Selection Method

Abstract

Understanding fetal health early can reduce risks to the pregnancy and the womb. Identifying correlations among factors influencing fetal well-being helps medical professionals clarify key impacts. Quantified relationships between features and labels also guide future research. This study focuses on three aspects: evaluating KNN model performance with and without ReliefF feature selection, analyzing the impact of feature removal, and assessing ReliefF's ability to identify critical features for fetal health classification.The research begins by framing fetal health classification as a supervised machine learning task using labeled datasets. A cardiotocographic dataset from the UCI Machine Learning Repository supports data collection. Initial analysis identifies data types and detects outliers, followed by preprocessing, feature selection, and KNN model training. Model testing uses metrics such as accuracy and recall. Results show the KNN model with ReliefF features achieves an accuracy of 0.896. Testing a pruned model by removing high-importance features slightly reduces accuracy to 0.866. These findings confirm ReliefF's effectiveness in identifying essential features and optimizing model efficiency without compromising quality. This study underscores ReliefF's role in improving KNN performance for fetal health classification.