Making Sense of Data: A Practical Guide to Exploratory Data Analysis and Data Mining

7.5 NEURAL NETWORKS

7.5.1 Overview

A neural network is a mathematical model that makes predictions based on a series of input descriptor variables. Like all prediction models, it uses a training set of examples to generate the model. This training set is used to generalize the relationships between the input descriptor variables and the output response variables. Once a neural network has been created, it can then be used to make predictions. The following sections describe what neural networks look like, how they learn and how they make predictions. An example is presented illustrating how neural networks can be optimized.

7.5.2 Neural Network Layers

A neural network comprises of a series of independent processors or nodes. These nodes are connected to other nodes and are organized into a series of layers as shown in Figure 7.25. In this example, each node is assigned a letter from A to L and organized into three layers. The input layer contains a set of nodes (A, B, C, D, E, F). Each node in the input layer corresponds to a numeric input descriptor variable. In this case, there are six input descriptor variables. The layer shown in black is the output layer containing nodes K and L. Each output node corresponds to an output response variable (two in this example). Between the input layer and the output layer is a hidden layer of nodes (G, H, I, J). In this example, there is just a single hidden layer comprised of four nodes. The number of hidden layers normally range from 0 ...