Wind energy has undergone rapid growth over the last two decades, demonstrating it to be one of the most promising renewable energy sources. However, since the average expected lifetime of a wind-turbine is around twenty-five years, the need to develop models to automatize damage detection, aiming for the extension of the lifespan of existing assets, is becoming high-priority. For this, continuous vibration-based structural health monitoring can be used. Nevertheless, environmental and operational variability leads to substantial dispersion in the modal properties of wind turbines potentially concealing damage-like changes. To reduce this variability, residual-based monitoring is used in this thesis, which investigates data-driven approaches with the purpose of developing prediction models that are able to forecast the natural frequencies of an onshore wind turbine tower under varying operational conditions.