Data Analysis with Open Source Tools

Chapter 18. Predictive Analytics

DATA ANALYSIS CAN TAKE MANY DIFFERENT FORMS—NOT ONLY IN THE TECHNIQUES THAT WE APPLY BUT ALSO in the kind of results that we ultimately achieve. Looking back over the material that we have covered so far, we see that the results obtained in Part I were mostly descriptive: we tried to figure out what the data was telling us and to describe it. In contrast, the results in Part II were primarily prescriptive: data was used as a guide for building models which could then be used to infer or prescribe phenomena, including effects that had not actually been observed yet. In this form of analysis, data is not used directly; instead it is used only indirectly to guide (and verify) our intuition when building models. Additionally, as I tried to stress in those chapters, we don’t just follow data blindly, but instead we try to develop an understanding of the processes that generate the data and to capture this understanding in the models we develop. The predictive power of such models derives from this understanding we develop by studying data and the circumstances in which it is generated.^[30]

In this chapter, we consider yet another way to use data—we can call it predictive, since the purpose will be to make predictions about future events. What is different is that now we try to make predictions directly from the data without necessarily forming the kind of conceptual model (and the associated deeper understanding of the problem domain) as discussed in ...