Correlation Detection and Judgment of Abnormal Individual Behavior Based on Unsupervised Learning

Abstract

In order to enable the Body Sensor Network (BSN) to detect the abnormal body state without the medical professional supervision, the method is proposed to find fuzzy association rules between daily exercise and physiological health data based on unsupervised learning. Firstly, the fuzzy clustering method is used to transform the numerical attributes to fuzzy categorical attributes. And then the frequent item set is calculated based on the support and confidence. The rules which reflect the behavior habit and laws are obtained from the final strong association rule through the minimum confidence screening. In order to avoid the useless rules and reduce the computational complexity, according to the characteristics of the exercise data, the frequent item set is divided into condition and conclusion groups. The abnormal state is detected by comparing the daily exercise data with the rules. The experiment with mass data had proven the validity of project.