Improving daily stochastic streamflow prediction: comparison of novel hybrid data-mining algorithms

In the current paper, efficiency of three new standalone data-mining algorithms [M5 Prime (M5P), Random Forest (RF), M5Rule (M5R)] and six novel hybrid bagging (BA-M5P, BA-RF BA-M5R) Attribute Selected Classifier (ASC-M5P, ASC-RF ASC-M5R) for streamflow prediction were assessed compared with an autoregressive integrated moving average (ARIMA) model as a benchmark. The models used precipitation (P) (Q) data from period 1979–2012 training validation (70% 30% data, respectively). Different input combinations prepared using both P Q different lag times. best combination proved to be that in which all (i.e. R – times). Overall, employing times more effective than only prediction. Although showed very good predictive power, BA-M5P outperformed other models.