Multibody dynamic models have been a fundamental tool for modelling the interaction between various members of a structure and its response to external loading. The complexity of such models depends on the number of elements, the nature of their interaction and the nature of external forces. To this end, wind turbines are one of the most challenging multibody systems which model interaction between rigid and flexible elements, involving their rotations and deformations. Further, wind turbines are subjected to highly stochastic external loading and operate under an uncertain loading environment. The authors demonstrate the application of machine learning, specifically LSTM models, to predict the deformation response of the wind turbine system. The most sensitive parameters governing the loads acting on the wind turbines are identified, and the samples are generated using the Sobol sequence. For each parameter combination, a turbulence box of the inflow wind is generated where multiple stochastic realisations are generated by varying the random seed of the generator to ensure that uncertainty due to turbulence pattern is captured. A novel feature identification method is used to discover the most informative subset of features. An LSTM model is developed to predict the blade deformations using these identified features. The LSTM model's performance is tested under varying levels of input uncertainties. The model performance is presented in Fig. 1 and 2. The developed model can act as a virtual sensor which captures the essential dynamics of wind turbine blades. Further, the model can be trained using measured data to act as a digital twin, which can be used for control and health monitoring applications.