DELIVERY METHOD PREDICTION USING BORUTA FEATURE SELECTION AND RANDOM FOREST ALGORITHM

Abstract

The subjective nature of delivery method decisions may affect the accuracy of maternal referrals. This study aims to develop an objective prediction model for delivery methods using the Random Forest algorithm and Boruta feature selection. The study utilized 503 medical records of pregnant women from Dr. Mohammad Hoesin Hospital Palembang. The Boruta algorithm was applied to select the most relevant features from 13 initial variables for training the Random Forest model. The results showed that Boruta successfully identified 12 important features, with preeclampsia, history of cesarean section, and mean arterial pressure as the primary clinical predictors. Evaluation of the Random Forest model achieved an accuracy of 94.00% with balanced performance and no indication of overfitting. In conclusion, the combination of Boruta and Random Forest methods produced an accurate delivery method prediction model with strong potential for application as a clinical decision support system.