The R Book

Neural Networks

These are computationally intensive methods for finding pattern in data sets that are so large, and contain so many explanatory variables, that standard methods such as multiple regression are impractical (they would simply take too long to plough through). The key feature of neural network models is that they contain a hidden layer: each node in the hidden layer receives information from each of many inputs, sums the inputs, adds a constant (the bias) then transforms the result using a fixed function. Neural networks can operate like multiple regressions when the outputs are continuous variables, or like classifications when the outputs are categorical. They are described in detail by Ripley (1996). Facilities for analysing neural networks in the MASS library.