To study the nonlinear characteristics of changes in the Earth's rotation rate, a comprehensive analysis of the nonlinear characteristics of the length of day (ΔLOD) observations reflecting changes in the Earth's rotation rate was conducted from multiple perspectives, including periodicity, chaos, and fractal, using the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), recursive quantitative analysis (RAQ), and Grassberger-Procaccia (GP) algorithms. The long-term high-accuracy ΔLOD observations from January 1, 1962 to December 31, 2023 were used as dataset published by the International Earth Rotation and Reference Systems Service, IERS) 14C04 series for fulfilling the comprehensive analysis and then achieve a reliable analyzed results. In the presented work, the emphasis was placed on comparing and analyzing whether there were significant differences in the ΔLOD characteristics before and after deducting the periodic or chaotic components of ΔLOD time series. The main conclusions are as follows. 1) The ΔLOD time series consists of the well-know trend components and many periodic components as well as chaotic components, and can therefore be characterized by obvious multi timescales, chaotic dynamics characteristics, and fractal structure. The characteristics were not noticed in previous research. 2) The period of the ΔLOD time series after deducting the chaotic components is exactly the same as the period of the original ΔLOD time series, implying that the chaotic components have no effects on reconstruction and analysis of the periodic components. 3) There is no significant difference in the chaotic characteristics between the original ΔLOD time series and its time series after deducting trend and periodic components, but the complexity of the fractal structure of the former is relatively stronger. Not only can this work provide a valuable reference to study the mechanism for changes in the Earth's rotation rate, but also model such rotation changes and then predict the chances in different timescales.