Real-time BigData and Predictive Analytical Architecture for healthcare application

Abstract

The amount of data produced within health informatics has grown to be quite vast. The large volume of data generated by various vital sign monitoring devices needs to be analysed in real time to alert the care providers about changes in a patients condition. Data processing in real time has complex challenges for the large volume of data. The real-time system should be able to collect millions of events per seconds and handle parallel processing to extract meaningful information efficiently. In our study, we have proposed a real-time BigData and Predictive Analytical Architecture for healthcare application. The proposed architecture comprises three phases: (1) collection of data, (2) offline data management and prediction model building and (3) real-time processing and actual prediction. We have used Apache Kafka, Apache Sqoop, Hadoop, MapReduce, Storm and logistic regression to predict an emergency condition. The proposed architecture can perform early detection of emergency in real time, and can analyse structured and unstructured data like Electronic Health Record (EHR) to perform offline analysis to predict patient’s risk for disease or readmission. We have evaluated prediction performance on different benchmark datasets to detect an emergency condition of any patient in real time and possibility of readmission.