Klasifikasi Kondisi Janin Berdasarkan Data Kardiotogram Menggunakan Algoritma Naive Bayes

Abstract

Fetal health during pregnancy is a crucial aspect that can be monitored through cardiotocography (CTG) data; however, manual interpretation of this data often encounters challenges due to class imbalance. This study aims to develop a fetal condition classification model using the Naive Bayes algorithm combined with the Synthetic Minority Over-sampling Technique (SMOTE) to address the disparity in class distribution. The CTG dataset, obtained from Kaggle, consists of 2,126 records categorized into three target classes: Normal, Suspect, and Pathological. Data processing followed the Knowledge Discovery in Databases (KDD) framework, including data selection, cleaning, normalization, splitting into four ratios (70:30, 80:20, 85:15, and 90:10), SMOTE application, and model evaluation using accuracy and F1-Macro metrics. The results showed that the 80:20 ratio yielded the highest accuracy at 79.81%, while the 90:10 ratio produced the highest F1-Macro score of 0.6788. These findings indicate that although accuracy remained relatively stable, the F1-Macro metric provided a better representation of performance across all classes, especially minority ones. The application of SMOTE proved effective in balancing class distribution and enhancing model sensitivity. This study serves as a foundational step in developing a more reliable and adaptive fetal condition classification system and highlights opportunities for further exploration of alternative algorithms and SMOTE parameter optimization.