Optimized Knowledge Management System for Integrated Databases Using Hybrid Semantic Graphs and Machine Learning Models in Predicting Kidney Failure with Urine Creatinine Levels

Abstract

Early detection of chronic kidney disease (CKD) is an increasingly critical priority in global healthcare systems due to its asymptomatic progression and high morbidity rates [10]. Despite the availability of extensive clinical data, such as laboratory results, medical histories, and physiological indicators, the predictive utility of these datasets remains limited by structural fragmentation and semantic inconsistencies across electronic health records [5]. These challenges significantly reduce the effectiveness of data-driven clinical decision support systems, which have shown promising outcomes in domains such as cardiovascular and chronic disease prediction [1], [3]. To overcome these limitations, this study proposes an optimized Knowledge Management System (KMS) that integrates a hybrid semantic graph architecture with machine learning techniques to improve early prediction of kidney failure risk, emphasizing urine creatinine levels as a key biomarker [8].

The system models heterogeneous clinical attributes—including demographics, comorbidity profiles, urine creatinine concentration, laboratory measurements, and estimated glomerular filtration rate (eGFR)—within a unified semantic graph constructed using standard medical ontologies such as SNOMED CT and ICD-10 [2], [9]. The hybrid semantic graph approach enhances interoperability by combining structured ontological relationships with numerical clinical data, allowing richer representation of medical knowledge and improved contextual interpretation of patient-level information [4], [11]. This graph-based integration not only captures complex attribute interdependencies but also enables multi-layer reasoning and scalable knowledge expansion across chronic disease domains [3], [11].

To evaluate predictive performance, four machine learning models—XGBoost [13], Gradient Boosting [12], Support Vector Machine (SVM) [14], and Multilayer Perceptron Neural Networks [15]—were trained on the graph-enriched dataset. Model performance was assessed using accuracy, F1-score, and Area Under the Curve (AUC). Experimental results demonstrate that integrating semantic graph structures yields a 15–25% performance improvement compared to conventional relational table-based models, consistent with trends observed in recent CKD-focused predictive analytics research [6], [7]. Among all predictors, urine creatinine exhibited the strongest predictive significance, reinforcing its role as a sensitive indicator of renal function decline and early-stage kidney injury [8].

Beyond predictive modeling, the developed KMS provides an extensible platform for clinical knowledge exploration, enabling users to visualize graph-based relationships, examine conceptual linkages, and interrogate high-dimensional clinical patterns. This semantic capability aligns with recent advances in ontology-driven analytics and knowledge graph–based clinical support systems, which have demonstrated improved interpretability and diagnostic insight across complex medical datasets [2], [9], [11]. The system’s modular architecture also allows for future incorporation of additional biomarkers, clinical entities, and disease pathways, supporting broader application to hypertension, diabetes, and cardiovascular disease [1], [3].

Overall, the findings indicate that hybrid semantic graph integration significantly enhances the predictive accuracy and interpretability of machine learning models for kidney failure risk assessment. The proposed KMS represents a substantive advancement in intelligent clinical informatics, demonstrating how semantic enrichment and data-driven analytics can synergistically improve early detection, clinical decision-making, and long-term disease management [1]–[15].