The use of accelerometers to assess physical activity in research studies has increased in recent years. Commercial devices represent a low cost, durable, and unobtrusive option for recording physical activity over extended time periods. Recent advances have allowed a shift in observation from the laboratory setting to the day to day lives of participants. Increases in observation time has led to subsequent increases in missing data due to non-wear, and a need for ways to address this. FDA, which regards the unit of observation to be a function over time or space, has clear applications in accelerometer research. Functional principal components analysis (FPCA), which aims to approximate functional observations by a combination of primary modes of variability, provides an effective strategy for estimating functional profiles in addition to data reduction. We propose a simulation study evaluating FPCA and a recent nonnegative decomposition approach (akin to FPCA) as methods for estimating the underlying functional profiles, and subsequently imputing plausible values for missing data. Evaluation of these methods on a real-world data set will also be discussed.