The tall building conceptual design phase is typically a time-consuming trial-and-error procedure that does not usually guarantee an optimal solution. An efficient optimization framework can be used to address complex challenges in layout planning. Yet the objective function evaluation process is computationally expensive, especially in cases where thousands of combinations of possible solutions should be investigated or time history analysis is required, as is the case in tall buildings with shear wall layouts. This conference paper presents a comprehensive comparative study on utilizing machine learning-based surrogate models within layout optimization frameworks for tall buildings. The focus is on evaluating the performance of various machine learning techniques in enhancing the efficiency and accuracy of layout optimization processes. The study explores the application of surrogate models, including random forests, support vector machines, Gaussian process regression, decision trees, and deep neural networks as crucial components of optimization frameworks. Each model is assessed for its ability to approximate the complex relationships governing shear wall layouts, which are crucial in resisting lateral loads in tall structures. The evaluation considers factors such as computational efficiency, predictive accuracy, and generalization capabilities across diverse design scenarios. The training samples are prepared using finite element analysis of multiple layouts based on the Latin-Hypercube sampling technique. This paper will focus on dynamic wind excitation, aiming to provide valuable insights into the strengths and limitations of different surrogate models, aiding structural engineers and architects in informed decision-making during the conceptual design phase.