Classification of Rest and Active Periods in Actigraphy Data Using PCA

In this paper we highlight a clustering algorithm for the purpose of identifying sleep and wake periods directly from actigraphy signals. The paper makes use of statistical Principal Component Analysis to identify periods of rest and activity. The aim of the proposed methodology is to develop a quick and efficient method to determine the sleep duration of an individual. In addition, a robust method that can identify sleep periods in the accelerometer data when duration, time of day varies by individual. A selected group of 10 individual's sensor data consisting of actigraphy from an accelerometer (3-axis), near body temperature, and lux sensors from a single GENEActiv watch worn on the non-dominant hand. The actigraphy of each individual was collected 24 hours a day for a period spanning 80 days. We highlight that a simple data preprocessing stage followed with a 2 phase clustering method provides results that align with previously validated methodologies.