Multiplayer online battle arena(MOBA) is currently one of the most popular genres of digital games around the world. With the development of E-sports, the impact of data analysis on MOBA games is increasing. The in-game variables like gold & experience are generally selected as indicators to evaluate the real-time game situations. However, there are few previous studies on forecasting game-evolving trends. To learn the trend information in time-series data, we propose a MOBA game trend prediction model based on the sequence-to-sequence structure, called MOBA-Trend. Firstly, we design a data scaling algorithm and use a low-pass filter to eliminate noise in the data. Then, the model takes both lineups and historical variable sequences as inputs. And the seq2seq structure with attention mechanism is used to forecast the future trends of gold & experience. Finally, we apply the model to Dota2, one of the most popular MOBA games. Experiments on a large number of match replays show that the model can effectively forecast the evolving trends.