Significant progress has been made in developing sleep staging methodologies; however, less attention has been devoted to the analysis of sleep architecture. Two critical aspects remain underexplored: the choice of binning window (i.e., grouping data into time intervals) and the statistical treatment of the interdependencies among sleep phases. While one-hour bins are commonly used, this choice is often based on convention rather than empirical justification. Additionally, sleep architecture data are typically expressed as proportions of time spent in each sleep phase, forming compositional datasets, non-negative values that represent parts of a whole and are constrained by a constant sum (e.g., 100%). Such data violate the assumptions of traditional statistical methods, yet their compositional nature is often overlooked, compromising analytical validity. In this study, we address two key methodological challenges in sleep architecture analysis: (1) determining the optimal binning window through a data-driven approach that balances information retention and noise reduction, and (2) applying isometric log-ratio (ILR) transformation to account for the compositional structure of the data, enabling the use of conventional statistical tests. By addressing these issues, we propose a more rigorous and interpretable framework for analyzing sleep architecture, aiming to enhance the accuracy and reproducibility of findings in sleep research.