This is a must have introductory book for the practitioner using data mining to build predictive models in industry. While it does have a few snippets of SAS code, it is a conceptual book that explains the "why" and the "how" of practical model building. (If you want SAS code buy "The Data Mining Cookbook" by Olivia Parr Rud.) It dispenses of with the antiquated notion of the "true" model of classical statistics and econometrics, and shows how to arrive at an acceptable model that yeilds good predictions. As practitioner's, this is what we care about most. Among other things, it gives good explanations of: (1) the EDA paradigm versus classical statistics (2) Tukey's bulging rule for transforming variables (3) variable selection, though there is no mention of clustering to eliminate redundant variables. It discusses some of the weaknesses of automatic variable selection methods (4) smoothed scatterplots and logit plots (5) decile analysis and using bootstrapping to derive confidence intervals for cum lift.
The book shows you how to use logistic regression, OLS, and CHAID to build predictive models. For those interested in Genetic modeling, it has a clearly written chapter on the subject that explains how genetic modeling can be used to create new variables that can have more information than either of the original variables.
While this book does not cover everything, and is definitely not the last word on the subject, it is a solid first word. In particular, the book does not cover splines, shrinkage techniques such as model averaging, ridge regression, ..etc. For treatments of these and similar advanced topics see Frank Harrell's "Regression Modeling Strategies" and Hastie, Tibsharani and Friedman's "Elements of Statistical Learning".
