An Ontology Alignment based on Machine learning for Integration of Patient Health Data

Abstract

The integration of patient data is crucial in healthcare informatics. It involves organizing and integrating heterogeneous health data from various Electronic Health Records (EHRs). Attribute alignment is a fundamental step in data integration. It involves mapping data attributes across different datasets. Most of the data maintained in EHRs does not follow standard terminologies in healthcare. Therefore, it becomes difficult to integrate patient health data from diverse data sources for generating historic medical records. The research work carried out overcomes this problem by developing a vital sign ontology using OpenEHR health standards. It helps to map the vital signs observations of the patients from its proprietary sources uniformly. The work also leverages the power of supervised learning algorithms to automate the mapping of different health datasets to the proposed ontology. The approach is evaluated on patient health datasets, considering both standard and non-standard datasets. The research work employs different machine learning algorithms, such as Support Vector Machine (SVM), Naive Bayes, Logistic Regression, k-nearest neighbor (KNN), AdaBoost, and Neural network, in order to evaluate the best algorithm for the proposed approach. The evaluation results conclude that Naive Bayes exhibits the highest accuracy, with minimum misclassification rate, in both the training and validation phases for automatically mapping the health datasets with the proposed ontology.