Machine Learning Approach of Semantic Mapping in Polystore Health Information Systems

Abstract

Health analysis and Information system use patient data from ubiquitous data sources for decision making. Limited adoption of health standards results in difficult data exchange and its reuse. The data integration from Polystore databases experience schema level conflict, which limits its sharing and reuse among other health organizations. The research work proposed an approach called Semantic Mapping of Observation Data for interoperability of Polystore Electronic Health record data sources. The proposed approach resolves schema level conflicts that occur while integrating patient EHR from multiple heterogenous data sources. The research work demonstrates SMOD on Blood Pressure data from standardized, non-standardized and streaming data sources. Its performance is accessed on widely used multiclass algorithms such as Support Vector Machine, K-Nearest Neighbor, Naive Bayes, Logistic Regression and Neural network. Results shows highest accuracy with Linear SVM in comparison with other classification algorithms. However, K-Nearest Neighbor and Naive Bayes performance is nearly close to SVM. The result is validated on Blood Pressure data taken from datasets of 3 different diseases such as Heart, Kidney and Diabetes. The validation results demonstrate that Naïve Bayes algorithm is used as best generalized algorithm in SMOD and is able to predict accurate mapping with other Blood Pressure datasets of different diseases.