Transport demand forecasting is essential when designing public transportation systems and introducing commercial transport modes. Although recent studies achieved high performances, they rely on a large number of datasets. In the real world situation, the acquisition of large data is often not easy. In extreme situations, it is impossible to obtain the data of the desired mode at all. In this study, we propose a method of synthesizing data of the desired transport mode using information learned from other modes which can be further used to predict the demand for transport modes without collected data. The proposed method learns the correlation between modes from multiple cities. We also utilize meta-learning techniques so that the model is trained to easily adapt to new tasks. We conducted extensive experiments using the transport mode data of various cities. Experiment results demonstrate that our model improves the performance by 1.93∼93.62% compared with existing methods.