OPTIMIZATION OF K VALUE IN THE K-NEAREST NEIGHBOR ALGORITHM FOR FETAL HEALTH CLASSIFICATION

Abstract

Fetal health classification is an essential step in supporting early diagnosis of fetal conditions and preventing pregnancy complications that can threaten the lives of both mother and fetus. The K-Nearest Neighbor (K-NN) algorithm is one of the commonly used methods for this classification due to its ability to recognize patterns from training data. However, the performance of K-NN is greatly influenced by the choice of the K value, which is the number of nearest neighbors considered in determining the class of new data. This study aims to optimize the K value in the K-NN algorithm to achieve the highest accuracy in fetal health classification. The data used consists of cardiotocography (CTG) examination results classified into three categories is Normal, Suspect, and Pathological. K values from 1 to 20 were tested to determine the optimal K value based on classification accuracy. The results indicate that the optimal K range is between 1 and 3, with the highest accuracy achieved at K = 1 and K = 2, both at 99.91%. In contrast, K values greater than 3 show a significant decrease in accuracy. Based on these findings, K = 3 is chosen as the optimal value to balance high accuracy with the model’s generalization capability. In conclusion, optimizing the K value in the K-NN algorithm can improve the accuracy of fetal health classification, potentially supporting more accurate and timely medical decision-making. These findings can be applied in developing a more reliable fetal health prediction system, thereby contributing to reducing the risk of maternal and fetal health complications