Model Complex Interactions with IBM SPSS Neural Networks

• Published in: SPSS pp. 325 - 353
• Format: Book Chapter
• Language: English
• Published: Indianapolis, Indiana John Wiley & Sons, Inc 04.04.2017
• Subjects: artificial neural networks; biological neurons; job classification; machine learning; multilayer perception (MLP); SPSS Algorithms
• ISBN: 9781119003557; 1119003555
• DOI: 10.1002/9781119183426.ch13

This chapter explores artificial neural networks as a technique available in the IBM SPSS Statistics Neural Networks module that uses a demonstration and two different case studies. Many discussions of neural nets discuss their similarity with biological neurons. The salient characteristic is that each biological neuron receives input signals from many other neurons via the receiving neuron's dendrites. The whole process is fascinating, and numerous books on machine learning discuss the process. Most discussions of neural nets that go into enough detail to mention the perceptron and the development of the multilayer perception (MLP) also mention this example. It can be easily found on the Internet, or in books on the subject. The chapter explains how the SPSS Algorithms Guide helps to clarify the exact nature of the implementation of MLP in SPSS. This is necessary because the broader literature will speak of many different flavors and variants. A change in one area of the network can have a ripple effect to other nodes in the network, with all kinds of compensation effects, sign changes, and canceling out being the result. Minority classification and a collection of variables representing job classification have been added. Most of the individual variables are significant. Interactions with these new variables could be considered, but the regression is already getting rather complex.