Sample Entropy

This chapter presents Sample Entropy (SampEn) alogorithm as an alternative method for entropy estimation in real world data. The chapter discusses the problems of approximate entropy (ApEn) algorithm and how SampEn addresses them. ApEn is optimally suited to measure the Gaussianity of a distribution and a process. Berg's theorem established that the maximum entropy for a random process with finite variance is attained by a Gaussian process. Thus ApEn values departing from the theoretical maximum indicate a lack of Gaussianity. SampEn can also be effectively used as a measure of Gaussianity though its maximum occurs for non-Gaussian random processes. When ApEn is used, modifications of ApEn handles zero and small number of matches can help minimize its bias and begin to approach the statistical stability of sample entropy. A formal implementation of SampEn is presented in this chapter. The practical issues of optimization of parameters and data filtering are also explained, which is followed by a discussion of the difficulties with short data sets and nonstationary data. Interpretation of entropy estimates is discussed and a direct comparison of ApEn and SampEn is also presented.