Investigation of the relative amplitude method in detecting early illness

Abstract

[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT AUTHOR'S REQUEST.] In this study, we examine the relative amplitude - a circadian rhythm measure, as a potential feature for early illness detection. We used data collected from sensor networks to analyze the daily activity level and sleep duration of the subjects. Using algorithms designed in earlier work, the data was preprocessed to remove unwanted data such as when a subject has a visitor, and to identify when the subject is away from home so as to compensate for lack of data for that duration. Using results from earlier work as a source of ground truth for separating normal days from abnormal days, various classification methods were investigated to separate the daily relative amplitude values into normal and abnormal days but the relative amplitude data appeared to be inseparable. A fuzzy inference system was then introduced with the maximum ten-hour activity level, time-home-alone, time-in-bed, restlessness, and relative amplitude as inputs. Although the fuzzy system identified some days as abnormal, they did not match the results obtained in the earlier work. Future work could include changing the activity level value that was substituted for the times a subject is away from home, and the fuzzy rules used could also be modified and made more robust.