Abstract—Reliable estimation of discharge is important in water resource planning and management, as well as in systems operation. This paper presents a rainfall runoff modelling approach using data mining techniques namely multi layer perceptron neural network and M5P-Model tree. Both models were developed, trained and verified for the discharge at Luvuvhu River, Mhinga gauging station. The relevant inputs into the models were selected by minimum Redundancy maximum relevance (mRMR) algorithm. The M5P Model Tree developed with 66% training set was realized to be the best model that predicted flow with a RMSE of 2.666, and a correlation coefficient of the observed and the predicted flow of 0.89. A MLP-ANN with 4 hidden nodes performed satisfactorily with RMSE ranging from 3.42 to 5.22. It is concluded that Model tree M5 predicts better than ANN-MLP, although it is quite sensitive to data splitting.
Index Terms—Stream flow prediction, MLP-ANN, M5P-model tree, mRMR.
The authors are with the Department of Civil Engineering, University of South Africa, Florida campus, South Africa (e-mail: email@example.com, firstname.lastname@example.org).
Cite: E. K. Onyari and F. M. Ilunga, "Application of MLP Neural Network and M5P Model Tree in Predicting Stream flow: A Case Study of Luvuvhu Catchment, South Africa," International Journal of Innovation, Management and Technology vol. 4, no. 1, pp. 11-15, 2013.