The Global Ocean Dissolved Oxygen Concentration Dataset (G4D-DOC) is a four-dimensional spatiotemporal product, derived from global Argo profiles and machine learning models. This dataset offers annual, seasonal, and monthly global ocean dissolved oxygen concentration information spanning the years from 2005 to 2022. It serves as a crucial data source for assessing global ocean ecological health, understanding global biogeochemical cycles, and conducting climate change research. The method involves four steps: 1) Data Collection and clean: The global BGC-Argo and Core-Argo profiles are gathered and systematically organized, and a rigorous quality control is performed to ensure the accuracy and reliability of the collected data; 2) Reconstruction Model: The ocean dissolved oxygen concentration reconstruction model (TSD-ML) is developed and honed through the utilization of the random forest model and the spatiotemporal deep partitioning method, and the global ocean dissolved oxygen concentration profiles, spanning the years 2005 to 2022, have been reconstructed; 3) Spatial Interpolation: The Spline Interpolation Method with Obstacles (SWP) has been employed to generate monthly grid data across 26 depth layers, maintaining a global resolution of 1°×1°. This was achieved by utilizing a global coastline vector and a seabed topography to create a spatial grid dataset of global ocean dissolved oxygen concentration; 4) Dataset Generation: The maximum value synthesis method is employed to generate seasonal and annual scale datasets derived from the monthly dataset.