Planning for water resources, reducing disasters, and controlling flooding can all be supported scientifically by runoff forecasts. Precise predictions are challenging to make in runoff prediction because of numerous uncertainties. The creation of a hybrid model (AOA-ELM), which is predicated on enhancing the extreme learning machine (ELM) by the application of the arithmetic optimization method (AOA), is the unique contribution of this paper. So as to train the AOA-ELM model on the Subarnarekha River in India, data from 2004 to 2023 (2004–2018) is used, whereas data from 2019 to 2023 is used for model prediction. On the same dataset, the conventional ELM was used as a comparison prediction model. The findings show that AOA-ELM performed better than standalone ELM in terms of coefficient of determination (R2), Nash–Sutcliffe efficiency (NSE), and root mean square error (RMSE). The findings show that AOA is a different training approach for choosing ELM parameters and that AOA-ELM may greatly enhance ELM's generalization performance for hydrologic time-series prediction. The suggested model can be a vital resource for accurate flood forecasting and prompt alerts.