Fetal health is a crucial aspect in reducing infant mortality rates, where cardiotocography (CTG) is used to monitor fetal condition through recordings of fetal heart rate and uterine contractions. However, manual interpretation of CTG data still faces challenges, particularly due to imbalanced class distribution. This study aims to develop a classification model for fetal conditions using the K-Nearest Neighbors (K-NN) algorithm combined with the Synthetic Minority Over-sampling Technique (SMOTE). The dataset used, sourced from Kaggle, consists of 2,126 CTG examinations categorized into three classes: Normal, Suspect, and Pathological. The data processing follows the Knowledge Discovery in Databases (KDD) process, including data selection, cleaning, normalization, splitting, balancing with SMOTE, and classification using K-NN. The model was evaluated using four training-testing split ratios (70:30, 80:20, 85:15, and 90:10) with accuracy and macro F1-score as metrics. The results indicate that the 85:15 split ratio achieved the highest accuracy of 89.7%, while the 90:10 ratio yielded the highest macro F1-score of 0.83. These findings suggest that the 85:15 ratio offers an optimal balance between model training and evaluation, whereas the highest F1-score at 90:10 reflects greater model sensitivity to minority classes. The combination of K-NN and SMOTE proved effective in addressing data imbalance and supports model stability in the overall classification process of fetal conditions.