Book Image

IBM SPSS Modeler Essentials

By : Jesus Salcedo, Keith McCormick
Book Image

IBM SPSS Modeler Essentials

By: Jesus Salcedo, Keith McCormick

Overview of this book

IBM SPSS Modeler allows users to quickly and efficiently use predictive analytics and gain insights from your data. With almost 25 years of history, Modeler is the most established and comprehensive Data Mining workbench available. Since it is popular in corporate settings, widely available in university settings, and highly compatible with all the latest technologies, it is the perfect way to start your Data Science and Machine Learning journey. This book takes a detailed, step-by-step approach to introducing data mining using the de facto standard process, CRISP-DM, and Modeler’s easy to learn “visual programming” style. You will learn how to read data into Modeler, assess data quality, prepare your data for modeling, find interesting patterns and relationships within your data, and export your predictions. Using a single case study throughout, this intentionally short and focused book sticks to the essentials. The authors have drawn upon their decades of teaching thousands of new users, to choose those aspects of Modeler that you should learn first, so that you get off to a good start using proven best practices. This book provides an overview of various popular data modeling techniques and presents a detailed case study of how to use CHAID, a decision tree model. Assessing a model’s performance is as important as building it; this book will also show you how to do that. Finally, you will see how you can score new data and export your predictions. By the end of this book, you will have a firm understanding of the basics of data mining and how to effectively use Modeler to build predictive models.
Table of Contents (19 chapters)
Title Page
About the Authors
About the Reviewer
Customer Feedback


Segmentation (as shown in the following screenshot) is very different from Classification. With segmentation or clustering, the main idea is that you do not believe that all the cases in the dataset are similar. That is, you believe there are distinct groups of people, and these groups should be looked at separately. Often, cluster analysis is used in marketing campaigns so that not all customers receive the same ads, but instead receive the appropriate ads. In this scenario, there is no dependent variable, only independent variables that are used to segment the cases.

K-means is the oldest of the techniques and has been popular and widely available for decades. It uses the distances between cases (on scale variables only) to determine similarity. Here, distance can be thought of quite literally—Euclidian distance is a common method. Cases whose values are close on the scale variables are grouped into clusters in an attempt to find homogeneous subsets. Determining how many clusters...