Conceptual design is crucial for designing offshore jacket substructures because it sets the direction for the entire design process. Nevertheless, conventional simulation-based optimization methods for jacket conceptual design face challenges, such as high computational costs and restricted optimization objectives. This paper proposes a data-driven method for offshore jacket conceptual design using machine learning (ML). First, a novel dataset of completed and under-construction jackets worldwide was established as the cornerstone of ML. The dataset comprised “in-action” data capturing key structural parameters of jackets and information on design boundary conditions. Subsequently, different features were comprehensively selected to identify and visualize their correlations for an interpretable data-driven design, ensuring the effectiveness of the dataset for training the ML models. Finally, random forest and eXtreme gradient boosting models were trained on the data from the selected feature subsets and then employed to predict individual jacket structural parameters. The predictive performance of the models indicates that the dataset and feature selection can capture the fundamental and shared characteristics of well-designed jackets, thereby improving the accuracy and efficiency of the conceptual design process. This study suggests the potential of a data-driven conceptual design for offshore jacket substructures.