The process of creating analytical datasets is simple in concept but hard in execution. You have probably created one already, whether you know it or not. Analytical datasets combine a bunch of useful features together into each record instance.
This is done for both data understanding purposes (think about Chapter 6, Getting to Know Your Data - Exploring IoT Data) and for ML purposes (think about Chapter 10, Data Science for IoT Analytics). The goal is to combine at least 80% of what an IoT analyst would need to answer questions for a particular subject into one table. The table is also constantly and automatically created for them. It is as simple as that.