The offshore wind sector is mature and has led to standardized design methods for offshore substructures. The conceptual design phase is critical for efficiency and cost-effectiveness and forms the basis for further design iterations. As turbine capacity increases, so does the complexity of offshore substructures, making design more challenging. However, the design process still relies on the expertise of the design engineers. These engineers rely heavily on their experience and intuition when designing, which can lead to biases due to limited information. To address this problem, Machine Learning (ML) techniques offer a promising way to improve the accuracy and efficiency of the conceptual design of offshore substructures. The current study is limited to the conceptual design of jacket substructures and was conducted on a self-developed global dataset of real jackets. The ML-based approach proposed in this study is capable of learning from existing data, recognizing intricate relationships between design variables, and potentially providing more accurate estimates for the initial conceptual design of offshore jacket substructures.