Many time series data have both time and space dimension which is known as space-time data. The objective of this research is to propose a hybrid Multivariate Generalized Space-Time Autoregressive Artificial Neural Network (MGSTAR- ANN) for handling both linear and nonlinear pattern in space-time data forecast. Air pollution data is used as a case study. The data consist of three pollutants, i.e. CO, NO2, and PM10 that were observed at three different locations, i.e. SUF 1, SUF 6, and SUF 7. RMSE (Root Mean Square Error) is used as an accuracy measurement for selecting the best model. The results show that a hybrid MGSTAR-ANN yield more accurate forecast than MGSTAR model. Moreover, these results are in line with one out of five major findings in the M4-Competition reported that the hybrid approach which utilized both statistical and Machine Learning features have more accurate result than the combination benchmark used to compare the submitted methods.