Predictive Analytics Techniques
- Statistics and Machine Learning
- The Impact of Big Data
- Supervised and Unsupervised Learning
- Linear Models and Linear Regression
- Generalized Linear Models
- Generalized Additive Models
- Logistic Regression
- Enhanced Regression
- Survival Analysis
- Decision Tree Learning
- Bayesian Methods
- Neural Networks and Deep Learning
- Support Vector Machines
- Ensemble Learning
- Automated Learning
We do not claim to review these methods exhaustively but present a general description of “families” of techniques, together with an explanation of the strengths and weaknesses for each family (see Exhibit 9.1).
Exhibit 9.1 Modern Analytics Framework
Many statistical techniques are useful for both prediction and explanation. Some techniques, however, such as Mixed Linear Models, are primarily useful for explanation, where the analyst seeks to assess the effect of one or more measures on another measure. The scope of this chapter does not include these techniques.
We begin the chapter with a brief discussion of two key “streams” of innovation in predictive analytics: statistics and machine learning. The distinction between these two streams is no longer as clear as it once was, because practitioners and advocates of each stream borrow from the other.
We also review the impact of Big Data. Some analysts argue that the Big Data phenomenon should have no impact on predictive analytics; these analysts argue that the core methods of predictive analytics do not change with the scale of the data. We disagree and therefore demonstrate specific ways in which Big Data can and will affect the techniques that analysts use.
In Chapter 7, “Analytic Use Cases,” we reviewed a number of use cases that require unsupervised learning techniques, such as segmentation, social network analysis, and text analytics. The unsupervised learning techniques required to support these use cases can also play a role in the predictive analytics workflow, so we include a brief discussion of these techniques.
The discussion of neural networks includes a brief overview of deep learning. Deep learning is a relatively recent innovation that has sparked new interest in applications for neural networks.
We close this chapter with a brief discussion of “meta-algorithms,” techniques to automate searches for an optimal model.