Introduction - Visual Data Mining: The VisMiner Approach - page 9

Model performance - Evaluate how well the model performs. If it is a
prediction model, how well does it predict? You can answer that question by
either comparing the model's performance to the performance of a random
guess, or by building multiple models and comparing the performance
of each.
Model understanding - Gain an understanding of how the model works.
Again, if it is a prediction model, you should ask questions such as: “What
input attributes contribute most to the prediction?” and “What is the nature
of that contribution?” For some attributes you may find a direct relationship,
while in others you may see an inverse relationship. Some of the relation-
ships may be linear, while others are non-linear. In addition, the contribu-
tions of one input may vary depending on the level of a second input. This is
referred to as variable interaction and is important to detect and understand.
Summary
In this chapter an overview of a methodology for conducting a data mining
analysis was presented. The methodology consists of four steps: initial data
exploration, dataset preparation, data mining modeler application, and model
evaluation. In the chapters that follow, readers will be guided through application
of themethodology using a visual tool for data mining - VisMiner. Chapter 2 uses
the visualizations and features of VisMiner to conduct the initial exploration and
do some dataset preparation. Chapter 3 introduces additional features of VisMiner
for dataset preparation not covered in Chapter 2. Chapters 4 through 7 introduce
the dataminingmethodologies available inVisMiner, with tutorials covering their
application and evaluation using VisMiner visualizations.