ABSTRACT

This chapter explores a few of the more popular advanced techniques; these include many machine learning methods, although new ones are being introduced continuously. Neural networks are one of the original machine learning methods that were developed to solve complex problems in computer science, such as character recognition and computer vision. The chapter shows few examples where neural networks can perform well and automatically detect the variable structure, whereas linear models would require more human intervention. It covers the simple case of a single regression or classification tree and then focuses on the more complex structures, which are combinations of many simple trees. There are many popular software packages for fitting tree models, such as SPSS, R, and Python. The chapter recommends using the tree-based approaches as a benchmark to determine whether a neural network has fitted the data completely or gotten stuck in the estimation processes and not truly found an optimal set of parameter estimates.