JMIR Publications

ABSTRACT

Background:

Diabetes is a pervasive chronic condition requiring early detection and effective management to mitigate severe complications. While traditional predictive models often use statistical or machine learning algorithms, these methods may lack the contextual understanding of medical data. Combining data-driven clustering techniques with ontology-based frameworks offers a promising avenue for more interpretable and effective prediction systems

Objective:

This study aims to enhance diabetes prediction accuracy by integrating improved clustering methods with ontology-based knowledge representation, enabling semantic reasoning and contextual insights.

Methods:

The proposed approach applies an optimized Louvain Community Detection algorithm to cluster patient data, revealing inherent patterns and groupings. Ontology development using Protégé enriches the dataset with semantic annotations, enabling meaningful analysis. A dual-phase predictive model leverages clustering and semantic querying to predict diabetes outcomes based on patient-specific parameters.

Results:

The integrated approach demonstrated superior performance on the Pima Indians Diabetes dataset. Experimental evaluations using an 80-20 dataset split and 10-fold cross-validation yielded accuracy scores of 97.56% and 96.74%, respectively. Other metrics, including precision, recall, and F1-score, confirmed the model's robustness and generalizability compared to existing methods.

Conclusions:

By combining clustering techniques with ontology-based frameworks, this study provides a robust, interpretable, and accurate approach to diabetes prediction. The integration of semantic reasoning enhances model adaptability and relevance, paving the way for advanced, personalized healthcare solutions.