Chapter 11. Neural Nets
In this chapter we describe neural nets, a flexible data-driven method that can be used for classification or prediction. Although considered a "blackbox" in terms of interpretability, neural nets have been highly successful in terms of predictive accuracy. We discuss the concepts of "nodes" and "layers" (input layers, output layers, and hidden layers) and how they connect to form the structure of a network. We then explain how a neural network is fitted to data using a numerical example. Because overfitting is a major danger with neural nets, we present a strategy for avoiding it. We describe the different parameters that a user must specify and explain the effect of each on the process. Finally, we discuss the usefulness of neural nets and their limitations.
Neural networks, also called artificial neural networks, are models for classification and prediction. The neural network is based on a model of biological activity in the brain, where neurons are interconnected and learn from experience. Neural networks mimic the way that human experts learn. The learning and memory properties of neural networks resemble the properties of human learning and memory, and they also have a capacity to generalize from particulars.
A number of successful applications have been reported in financial applications [see Trippi and Turban, (1996)] such as bankruptcy predictions, currency market trading, picking stocks and commodity trading, detecting fraud in credit ...