This paper investigates rate-splitting multiple access (RSMA) networks assisted by aerial intelligent surfaces (AIRS) by employing deep-learning approaches to solve trajectory prob- lems for unmanned aerial vehicles (UAVs). Specifically, two models for predicting positions using long-short term memory (LSTM) and Transformers are developed. Training results show that both proposed frameworks can capture temporal features to determine the UAV’s position for tracking user mobility. However, simulation results indicate that the proposed Transformer-based model demonstrates robustness against variations in user loca- tions, providing superior prediction accuracy and consequently yielding higher performance gains in terms of sum rate when compared with the LSTM-based model. Additionally, it is demon- strated that the AIRS-RSMA scheme outperforms AIRS-NOMA systems due to its ability to effectively handle residual successive interference cancellation (SIC) errors.