Mining health knowledge graph for health risk prediction

Nowadays classification models have been widely adopted in healthcare, aiming at supporting practitioners for disease diagnosis and human error reduction. The challenge is utilising effective methods to mine real-world data in the medical domain, as many different models have been proposed with varying results. A large number of researchers focus on the diversity problem of real-time data sets in classification models. Some previous works developed methods comprising of homogeneous graphs for knowledge representation and then knowledge discovery. However, such approaches are weak in discovering different relationships among elements. In this paper, we propose an innovative classification model for knowledge discovery from patients’ personal health repositories. The model discovers medical domain knowledge from the massive data in the National Health and Nutrition Examination Survey (NHANES). The knowledge is conceptualised in a heterogeneous knowledge graph. On the basis of the model, an innovative method is developed to help uncover potential diseases suffered by people and, furthermore, to classify patients’ health risk. The proposed model is evaluated by comparison to a baseline model also built on the NHANES data set in an empirical experiment. The performance of proposed model is promising. The paper makes significant contributions to the advancement of knowledge in data mining with an innovative classification model specifically crafted for domain-based data. In addition, by accessing the patterns of various observations, the research contributes to the work of practitioners by providing a multifaceted understanding of individual and public health.