In recent decades, the potential alternatives to fossil fuel have been driven by concern over its environmental pollution and high cost. At present, the utilization of wind turbine technology to harness renewable wind energy has become widespread. The structure of vertical-axis wind turbines is simple, economical, and does not require yaw devices, suitable for urban areas. However, in the onset region of the atmospheric boundary layer, these turbines encounter continuous unsteady airflow, the central shaft encounters substantial oscillations and results in system failure. Conventional experimental analysis and Computational Fluid Dynamics (CFD) entail a significant number of iterations, which prove to be both costly and time-consuming. In contrast, deep learning offers a viable alternative by approximating the input–output mapping without the need to solve intricate physical equations. In this paper, two deep learning models, MLP and LSTM, were designed and trained using experimental data to predict shaft vibration at various pitch angles. The LSTM model performed better, achieving accurate predictions with less Root Mean Squared Error (RMSE). This model offers to be more efficient and less time-consuming alternative to traditional wind tunnel tests and numerical simulations for determining blade displacement in wind turbines.