SPSS Statistics for Data Analysis and Visualization

Chapter 13 Model Complex Interactions with IBM SPSS Neural Networks

In this chapter, we are going to use a demonstration and two different case studies to explore artificial neural networks, as a technique available in the IBM SPSS Statistics Neural Networks module. Along the way, we are also going to pick up on some general data mining skills. In particular, we will learn how to create and use a partition variable in SPSS Statistics even though there is no dedicated menu for creating partitions. We will also discuss some theory, but focus on what is necessary to know regarding the why and how of the technique. A key strategy for that will be to use linear regression as a point of comparison.

We could explore a number of other related topics, but we choose not to pursue them here, with the goal of achieving sufficient depth in the topics that we do explore. We will discuss only multilayer perceptrons. We will not discuss the theory or the practice of radial basis functions, which is another kind of artificial neural network (ANN). Both are available types of neural nets in SPSS Statistics. We make the choice we do because it is the more commonly used and because it is more closely tied to the earliest incarnations of the technique. It will make it easier to make the transition between the history of the technique and the examples we will run in SPSS. Also, we will spend minimal time on the “tuning” of models. Changing settings in order to boost performance is of value, but ...