Modeling Uncertainty and Interdependencies in Medical Risk Factors: a Neutrosophic Approach to Complex Health System

Abstract

Cancer remains an intimidating challenge, a complex and heterogeneous group of diseases which has several root causes becomes a challenging deadly disease to the provinces of medicine and science. In recent years, from the younger generation to older, invariably in all the age groups, the cancer affected persons can be seen in common. Lung cancer, particularly Non-small cell lung cancer is one of the leading causes of cancer- related deaths globally. Lung cancer is often diagnosed only at advanced stages, where there are limited treatment options available. This paper delves into recent advancements in understanding and analysing the multifaceted factors contributing to lung cancer, emphasizing its molecular complexity. This study integrates Neutrosophic Cognitive Maps (NCMs) with machine learning models to predict lung cancer with high precision and recall. Logistic Regression, KNN, and XGBoost demonstrate robust performance across metrics, while Random Forest excels in sensitivity. The findings affirm the approach’s potential for accurate, personalized lung cancer diagnosis, with scope for enhancement through feature engineering and optimization

Introduction: Neutrosophic Cognitive Maps (NCMs) are a combination of Neutrosophy and cognitive maps, providing a framework to model and analyse complex systems with uncertain, indeterminate, or incomplete information. On the other hand, Computational Intelligence which handles uncertainty, imprecision and approximate reasoning which helps in solving complex problems. This field encompasses various techniques, including fuzzy logic, neural networks, machine learning, and other approaches that emulate human-like intelligence for decision-making and problem-solving in the presence of uncertainty. Researchers have investigated hybrid methodologies that merge Neutrosophic cognitive maps and soft computing in order to capitalise and leverage their respective strengths.

Cancer is a complex disease occurred by abnormal cell behaviour due to genetic mutations, epigenetic alterations, environmental factors, and lifestyle choices. Early detection and accurate classification of lung nodules are crucial for effective treatment and improved survival rates. While traditional methods have provided valuable insights to understand the individual molecular pathways involved in  cancer , they often fail to capture the big picture of how various other factors contribute to cause, develop and spread cancer. Neutrosophic Cognitive Maps (NCMs) offer a promising methodology to model the inherent uncertainty and complexity of cancer by incorporating fuzzy, contradictory, and indeterminate information. By integrating NCMs with advanced optimization algorithms, this study aims to unravel the detailed network of interactions governing cancer development across multiple scales, from molecular signalling pathways to environmental exposures.

Objectives: This paper aims to address indeterminacy, which is especially critical in medical fields where relationships often involve inherent uncertainty due to variability in patient data, environmental factors, and complex interactions, highlighting the hierarchical importance and interdependencies among these factors.

Methods: By introducing a hybrid model that integrates Neutrosophic Cognitive Maps and Machine Learning models to tackle uncertainty and inconsistency in cancer-related factors, addressing a critical issue in real-world medicine. Its capacity to address the inherent uncertainties and complexities in these domains fosters innovative problem-solving approaches and enhances the understanding of multifaceted phenomena. This hybrid method offers distinct disease diagnosis capabilities, facilitating physicians in patient treatment.

Results: The analysis validates the proposed methodology for lung cancer prediction by demonstrating that the integration of Neutrosophic Cognitive Maps (NCMs) and Machine Learning models can effectively predict lung cancer with high precision and recall. Logistic Regression, KNN, and XGBoost emerge as the most robust models, providing a balanced performance across all metrics, while Random Forest proves valuable in scenarios where sensitivity is paramount. The ROC-AUC scores, though moderate, suggest room for improvement in capturing the nuances of the prediction task, potentially through advanced feature engineering or hyperparameter optimization.

Conclusions: These results affirm the validity of the process and the applicability of the proposed approach in personalized diagnostic frameworks, highlighting its potential for real-world implementation in lung cancer prediction and management.