Radial Basis Function and Multilayer Perceptron neural networks for sea water optically active parameter estimation in case II waters: A comparison

Estimating the concentration of optically active constituents of sea water by Takagi–Sugeno models with quadratic rule consequents

“…[3,16], respectively. RBFs have been extensively used as non-linear regression models because of their simple topological structure and speed of the training phase.…”

“…Finally, we show that the second-order TS models achieve approximation performance considerably superior to the RBF networks and the MLP networks proposed in Refs. [3,16], and applied to the same simulated data.…”

Artificial neural network modeling of the water quality index for Kinta River (Malaysia) using water quality variables as predictors

“…However, the logistic sigmoid and hyperbolic tangent functions are the functions most commonly used with the MLP neural networks (Dawson and Wilby, 2001;Maier and Dandy, 2000) while the Gaussian function is the one most commonly used with the RBF neural networks (Corsini et al, 2003;Dawson and Wilby, 2001). The logistic sigmoid function is particularly appealing when the raw data have outliers because this function reduces the effects of extreme input values on the performance of the network and hence extreme values will have no extreme effects on the network outputs (Hill et al, 1994).…”

“…To efficiently find a reasonable learning rate, networks are trained and the outputs of the validation dataset are evaluated (Bruzzone et al, 2004;Corsini et al, 2003). During this process, the ANN model is trained and continually optimized against the cross-validation dataset so that the network structure can be assessed by using the error values produced during training on the cross-validation data.…”

Snow cover thickness estimation using radial basis function networks

“…Their approximations are smooth and continuous, and their accuracy increases with increasing numbers of nodes in the hidden layers. The benefits and limitations of MLP networks have become increasingly apparent and the results of comparative studies in diversified domains are now available (Corsini et al, 2003;Jayawardena et al, 1997). MLP is highly nonlinear in its parameters.…”