Handling Errors in eHealth Sensors Using Interval Mapping and Fuzzy Modeling

Abstract

EHealth sensors are connected through an Internet of Things infrastructure and the vital sign measurements are sent to cloud for real-time diagnosis by a remotely located physician. We use a cloud-based virtual sensing technology for collecting the vital signs of the patients remotely for smart diagnosis. Here the challenge is that the uncertainty generated from underlying physical sensor layer is propagated to the virtual sensors in cloud inflicting the diagnosis process. This article focuses on handling this challenge by introducing a new approach, Interval Mapping Technique (IMT), which is used to indicate the presence of any inaccuracy of devices and random errors at real-time. The novelty of this technique lies in the consideration that the Systolic and Diastolic BP of a person fall within certain ranges and the knowledge about these two ranges or intervals together can be used to get rid of random errors. A fuzzy modeling based approach is also proposed in this article. Fuzzy linguistic terms are used to determine the systolic and diastolic intervals of a person and a rule base is developed to find any real time error. Primarily, the techniques are used for error reduction in blood pressure measurements of Hypertensive, Pre-hypertensive and Normotensive patients. However, these methodologies can be extended for other vitals as well. These two techniques are compared with other standard techniques, like linear regression and random forest. Results show that IMT generates 3.69% less error for Systolic and 7.24% less error for Diastolic BP compared to linear regression model, 1.91% less error for Systolic and 7.18% less error for Diastolic BP compared to random forest model.