Large volume of multi-structured and low-latency patient data are generated in healthcare services, which is challenging task to process and analyze within the Service Level Agreement (SLA). In this paper, a Parallel Semi-Naive Bayes (PSNB) based probabilistic method is used to process the healthcare big data in cloud for future health condition prediction. In order to improve the accuracy of PSNB method, a Modified Conjunctive Attribute (MCA) algorithm is proposed for reducing the dimension. Emergency condition of the patient is considered by setting a global priority among the patients and an Optimal Data Distribution (ODD) algorithm is proposed to position both batch and streaming patient data into the Spark nodes. Further, a Dynamic Job Scheduling (DJS) algorithm is designed to schedule the jobs efficiently to the most suitable nodes for processing the data taking SLA into account. Our proposed PSNB algorithm provides better accuracy of 87.8% for both batch and streaming data, which is 12.8% higher than the original NaiveBayes (NB) algorithm and can conveniently be employed in various patient monitoring applications.

Introduction

Digital revolution such as Internet of Things (IoT) [1], Wireless Body Networks (WBNs) [2], Big Data [3] and Cloud Computing [4] enables the day-to-day living style easier and better. Big Data deals with extremely large data sets having four different characteristics including Volume, Variety, Velocity and Veracity. Besides, ceaseless streams of healthcare data are generated in large volume by ubiquitous smart devices such as smart phone, pulse oximeters and body sensors on real-time patient monitoring. Under the existing solution methods, it is very difficult to analyze and process both streaming and batch data together in a single platform within the deadline. As a result, the first problem is how to reduce the dimension of the health parameter for better accuracy. The second problem is how to find the dependencies among the healthcare parameters and priority of the patients based on the influential parameters. Thirdly, which appropriate method can be used for analysis and processing of those multi-structured, low-latency patient data with higher accuracy and efficiency. By considering above-mentioned issues, Big Data analysis, and processing are two major challenges in the sizable healthcare industry. In data analysis, various classification [5], clustering [6] and predictive analytic [7] algorithms are used based on the input and output data sets. However, many of those tools are outdated [8] as they are unable to handle large volume of multi-structured healthcare data sets. Specifically, in healthcare, the patient data are not only large in volume but also are generated with a tremendous speed, which requires an advanced platform for both analysis and processing. Also, the health condition of a patient is always related with some uncertain factors based on the clinical parameters. For probabilistic approach, Naive Bayes (NB) [9] is the best and most popular algorithm due to its efficiency. However, NB algorithm can be applied only on the independent data sets, which is not suitable for healthcare big data as most of the data have dependency among multiple parameters. Hence, the Semi-Naive Bayes (SNB) [10] algorithm can be used, which allows certain degree of dependency on input parameters. In healthcare applications, missing of any SLA [11] has highest impact on emergency patient data analysis due to the severity of the disease. The emergence of Big Data demands a distributed environment with parallel and fast computation.