Abstract

p1

Contents

p3

1 Error Rates in Discriminant Analysis

p1

1.1 Introduction

p2

1.2 Prediction error and bias-correction

p7

2 Comparison of Procedures and Bootstrap

p13

2.1 Introduction

p14

2.2 Bootstrap bias estimation

p15

2.3 Parametric bootstrap method

p20

2.4 Jackknife and Cross-validation

p32

2.5 Mean squared error

p35

3 Numerical Comparisons

p41

3.1 Introduction

p42

3.2 Characteristics of population

p43

3.3 Numerical experiments with LDF

p44

3.4 Numerical experiments with LDF and QDF

p54

3.5 Application to medical diagnosis

p58

4 Variable Selection Procedures

p61

4.1 Introduction

p62

4.2 Several criterions

p62

4.3 Monte Carlo experiment

p66

4.4 Example of medical diagnosis

p72

5 Shrinkage estimator

p77

5.1 Introduction

p78

5.2 Linear discriminant function

p79

5.3 Mean squared error

p81

5.4 Actual error rates

p89

5.5 Numerical results

p92

5.6 Discussion

p97

6 Efficient Bootstrap Method

p99

6.1 Introduction

p100

6.2 Bias of the apparent error rate

p101

6.3 Variance reduction method

p103

6.4 Monte Carlo study

p106

6.5 Discussion

p115

7 Conclusion

p117

Acknowledgments

p126

References

p127