Data Mining and Business Analytics with R

JOHANNES LEDOLTER, PhD, is Professor in both the Department of Management Sciences and the Department of Statistics and Actuarial Science at the University of Iowa. He is a Fellow of the American Statistical Association and the American Society for Quality, and an Elected Member of the International Statistical Institute. Dr. Ledolter is the coauthor of Statistical Methods for Forecasting, Achieving Quality Through Continual Improvement, and Statistical Quality Control: Strategies and Tools for Continual Improvement, all published by Wiley.

"I first taught a Ph.D. level course in business applications of data mining 10 years ago. I regularly search the web, looking for business-oriented data mining books, and this is the first one I have found that is suitable for an MS in business analytics. I plan to use it. Anyone who teaches such a class and is inclined toward R should consider this text." (Journal of the American Statistical Association, 1 January 2014)

• Preface ixAcknowledgments xi1. Introduction 1Reference 62. Processing the Information and Getting to Know Your Data 72.1 Example 1: 2006 Birth Data 72.2 Example 2: Alumni Donations 172.3 Example 3: Orange Juice 31References 393. Standard Linear Regression 403.1 Estimation in R 433.2 Example 1: Fuel Efficiency of Automobiles 433.3 Example 2: Toyota Used-Car Prices 47Appendix 3.A The Effects of Model Overfitting on the Average Mean Square Error of the Regression Prediction 53References 544. Local Polynomial Regression: a Nonparametric Regression Approach 554.1 Model Selection 564.2 Application to Density Estimation and the Smoothing of Histograms 584.3 Extension to the Multiple Regression Model 584.4 Examples and Software 58References 655. Importance of Parsimony in Statistical Modeling 675.1 How Do We Guard Against False Discovery 67References 706. Penalty-Based Variable Selection in Regression Models with Many Parameters (LASSO) 716.1 Example 1: Prostate Cancer 746.2 Example 2: Orange Juice 78References 827. Logistic Regression 837.1 Building a Linear Model for Binary Response Data 837.2 Interpretation of the Regression Coefficients in a Logistic Regression Model 857.3 Statistical Inference 857.4 Classification of New Cases 867.5 Estimation in R 877.6 Example 1: Death Penalty Data 877.7 Example 2: Delayed Airplanes 927.8 Example 3: Loan Acceptance 1007.9 Example 4: German Credit Data 103References 1078. Binary Classification, Probabilities, and Evaluating Classification Performance 1088.1 Binary Classification 1088.2 Using Probabilities to Make Decisions 1088.3 Sensitivity and Specificity 1098.4 Example: German Credit Data 1099. Classification Using a Nearest Neighbor Analysis 1159.1 The k-Nearest Neighbor Algorithm 1169.2 Example 1: Forensic Glass 1179.3 Example 2: German Credit Data 122Reference 12510. The Na¨ýve Bayesian Analysis: a Model for Predicting a Categorical Response from Mostly CategoricalPredictor Variables 12610.1 Example: Delayed Airplanes 127Reference 13111. Multinomial Logistic Regression 13211.1 Computer Software 13411.2 Example 1: Forensic Glass 13411.3 Example 2: Forensic Glass Revisited 141Appendix 11.A Specification of a Simple Triplet Matrix 147References 14912. More on Classification and a Discussion on Discriminant Analysis 15012.1 Fisher’s Linear Discriminant Function 15312.2 Example 1: German Credit Data 15412.3 Example 2: Fisher Iris Data 15612.4 Example 3: Forensic Glass Data 15712.5 Example 4: MBA Admission Data 159Reference 16013. Decision Trees 16113.1 Example 1: Prostate Cancer 16713.2 Example 2: Motorcycle Acceleration 17913.3 Example 3: Fisher Iris Data Revisited 18214. Further Discussion on Regression and Classification Trees, Computer Software, and Other Useful Classification Methods 18514.1 R Packages for Tree Construction 18514.2 Chi-Square Automatic Interaction Detection (CHAID) 18614.3 Ensemble Methods: Bagging, Boosting, and Random Forests 18814.4 Support Vector Machines (SVM) 19214.5 Neural Networks 19214.6 The R Package Rattle: A Useful Graphical User Interface for Data Mining 193References 19515. Clustering 19615.1 k-Means Clustering 19615.2 Another Way to Look at Clustering: Applying the Expectation-Maximization (EM) Algorithm to Mixtures of Normal Distributions 20415.3 Hierarchical Clustering Procedures 212References 21916. Market Basket Analysis: Association Rules and Lift 22016.1 Example 1: Online Radio 22216.2 Example 2: Predicting Income 227References 23417. Dimension Reduction: Factor Models and Principal Components 23517.1 Example 1: European Protein Consumption 23817.2 Example 2: Monthly US Unemployment Rates 24318. Reducing the Dimension in Regressions with Multicollinear Inputs: Principal Components Regression and Partial Least Squares 24718.1 Three Examples 249References 25719. Text as Data: Text Mining and Sentiment Analysis 25819.1 Inverse Multinomial Logistic Regression 25919.2 Example 1: Restaurant Reviews 26119.3 Example 2: Political Sentiment 266Appendix 19.A Relationship Between the Gentzkow Shapiro Estimate of “Slant” and Partial Least Squares 268References 27120. Network Data 27220.1 Example 1: Marriage and Power in Fifteenth Century Florence 27420.2 Example 2: Connections in a Friendship Network 278References 292Appendix A: Exercises 293Exercise 1 294Exercise 2 294Exercise 3 296Exercise 4 298Exercise 5 299Exercise 6 300Exercise 7 301Appendix B: References 338Index 341