Despite increased advancements of medical technology,influenza continues to pose tremendous threats to public health today.Accurate real-time prediction and immediate response are essential to avoid the repercussions of an influenza epidemic.In this paper,the influenza-like illness(ILI)data was adopted from the Centers for Disease Control and Prevention in the United States to predict influenza prevalence.In order to accurately predict influenza prevalence at a region level,the improved multi-verse optimizer(IMVO)by updating the traveling distance rate(TDR)was obtained for optimizing the parameters of Elman neural network(ERNN),whose model is written as IMVO-ERNN.By comparison,the results show that IMVO-ERNN outperforms MLR,ERNN and MVO-ERNN and is capable of predicting real-time regional estimates of influenza outbreaks in the United States.