For real-time precise positioning, the international GNSS service (IGS) real-time service (RTS) can be used to reduce GNSS orbit and clock errors. Since the IGS RTS is provided over the internet, it may be affected by intermittent data loss. We propose a prediction method to provide continuous RTS correction data during data loss period. An autoregressive moving average (ARMA) model is used to predict the RTS time series while machine learning algorithms are used to compute ARMA coefficients. Several machine learning algorithms, including neural network (NN) and genetic algorithm (GA), are tested with the ARMA model. One hour length of RTS orbit and clock correction data is used for training process, and then prediction process is performed up to 30 minutes. The machine learning prediction accuracy of the ARMA-machine learning algorithms are compared with a conventional ARMA algorithm. The machine learning algorithms outperforms the conventional ARMA algorithm. User positioning errors are computed by applying the predicted RTS data to GNSS signals, and then the positioning error levels are compared.