A Neutrosophic Random Forest Approach for Preeclamptic Risk Prediction with Uncertainty Quantification

Abstract

This study presents a novel integration of Random Forest with neutrosophic logic to improve preeclampsia risk prediction while quantifying prediction uncertainty. Using clinical data from 352 patients, the model achieved 72.73% accuracy with high sensitivity (0.898) in identifying control cases, though with lower specificity (0.235) for preeclampsia detection. Key predictors identified were birthweight and hypertension his- tory, aligning with clinical knowledge. The neutrosophic framework successfully categorized predictions into truth (T), indeterminacy (I), and falsity (F) components, revealing that 90% confidence predictions showed T = 0.9 while uncertain cases (0.5 ≤ p < 0.9) demonstrated elevated indeterminacy (I = 0.3). The main contributions include: 1) an interpretable uncertainty quantification method for clinical predictions, 2) validation of key risk factors through feature importance analysis, and 3) a practical framework for identifying cases requiring additional clinical evaluation. This approach demonstrates significant potential for enhancing decision-making in maternal healthcare.