Real-Time Meta Learning Approach for Mobile Healthcare

Abstract

This paper presents a novel real-time meta learning approach for predicting health risk of a patient under observation on a smartphone. While making health predictions, consideration of patient’s history and real-time trend of signal behavior is very important. This paper discusses the real-time healthcare system which learns the trend of various physiological signals with newly designed real-time stream mining algorithm PARC-Stream. It makes a health risk prediction on the fly using combination of both historical and dynamic risk rule base of patient. This meta Learning approach increases the chance of accurate risk prediction. Our experimental results proved that our novel meta learning approach used for health risk prediction gives a high prediction accuracy of 99% over other methods of using only single historical rule base or thresholds.