Several recent research efforts aim in providing a data-driven characterization of the response of structural systems, adopting machine learning techniques [1, 2]. Goal of those efforts is to minimize the computational cost accompanied by conventional numerical approaches and offer a fast, as well as accurate prediction of the response. In this work a numerical framework is proposed, towards evaluating the non-linear dynamic response of multi-storey steel buildings using machine learning techniques. Datasets consisting of input-output parameters characterizing the structural properties, geometry and seismic response are derived using classical non-linear time history analysis. Data sampling was based on the concept of analysing a range of steel buildings with heights between 2 to 10 storeys and number of bays between 2 to 10. Emphasis is given in critical elements that determine the dynamic response, including the influence of the geometry, potential use of shear plate walls and their failure as well as the lateral drifts which are obtained for the investigated structures. The capacity of traditional Artificial Neural Networks to predict the dynamic response of steel multi-storey buildings using Levenberg-Marquardt or Bayesian optimization algorithms was investigated. Therefore, the conducted research will highlight the steps of the methodology and will recommend appropriate machine learning and/or deep learning tools, that will successfully predict the dynamic response of the structures. REFERENCES [1] G.A. Drosopoulos and G.E. Stavroulakis. Non-linear Mechanics for Composite Heterogeneous Structures. CRC Press, Taylor and Francis, 2022. [2] S.M. Motsa, G.E. Stavroulakis and G.A. Drosopoulos. A data-driven, machine learning scheme used to predict the structural response of masonry arches. Engineering Structures, 296:116912, 2023.