A Comparative Study of MLR, KNN, ANN and ANFIS Models with Wavelet Transform in Monthly Stream Flow Prediction

Abstract

Reliable and precise prediction of the rivers flow is a major concern in hydrologic and water resources analysis. In this study, multi-linear regression (MLR) as a statistical method, artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) as non-linear ones and K-nearest neighbors (KNN) as a non-parametric regression method are applied to predict the monthly flow in the St. Clair River between the US and Canada. In the developed methods, six scenarios for input combinations are defined in order to study the effect of different input data on the outcomes. Performances of the models are evaluated using statistical indices as the performance criteria. Results obtained show that adding lag times of flow, temperature and precipitation to the inputs improve the accuracy of the predictions significantly. For a further investigation, the aforementioned models are coupled with wavelet transform. Using the wavelet transform improves the values of Nash-Sutcliff coefficient to 0.907, 0.930, 0.923, and 0.847 from 0.340, 0.404, 0.376 and 0.419 respectively, by coupling it with MLR, ANN, ANFIS, and KNN models.