Neutrosophic Uncertainty Modeling Along With Machine Learning For Breast Cancer Outcomes: A Hybrid Intelligent Framework for Medical Prognosis

Resumen

Strong prognostic techniques that can handle the inherent ambiguities, uncertainties, and inconsistent data common in clinical practice are essential for managing breast cancer. This paper introduces a novel hybrid computational framework that combines the predictive power of machine learning (ML) with the mathematical formalism of Neutrosophic Sets for uncertainty quantification. We propose that clinical data is not just imprecise but is essentially defined by three separate dimensions: falsity (contradictory evidence), indeterminacy (ambiguous or absent evidence), and truth (supportive evidence). In order to clearly characterize these aspects of uncertainty, our method first converts unprocessed clinical data into a neutrosophic feature space. The processed data is used to train and evaluate eight ML models (Support Vector Machine (SVM), k-Nearest Neighbors (K-NN), XGBoost, Logistic Regression, Random Forest, Neural Networks, Naïve Bayes, and Decision Trees) across five critical prognostic tasks: Diagnosis (Benign/Malignant), Recurrence, Chemotherapy Recommendation, Mortality and Survival. Several models achieved flawless performance (100% accuracy, precision, recall, and F1-score) in deterministic tasks such as Chemotherapy Recommendation and near-perfect diagnosis (SVM Accuracy: 97.37%, F1-Score: 97.93%), demonstrating the remarkable effectiveness of the framework. With the top F1-scores at 51.85% and 26.67%, respectively, the model outputs accurately reflect the inherent difficulties for challenging tasks like recurrence and survival prediction.