Pattern Recognition Approach to Data Interpretation

An attempt is made in this book to give scientists a detailed working knowledge of the powerful mathematical tools available to aid in data interpretation, especially when con­ fronted with large data sets incorporating many parameters. A minimal amount of com­ puter knowledge is necessary for successful applications, and we have tried conscien­ tiously to provide this in the appropriate sections and references. Scientific data are now being produced at rates not believed possible ten years ago. A major goal in any sci­ entific investigation should be to obtain a critical evaluation of the data generated in a set of experiments in order to extract whatever useful scientific information may be present. Very often, the large number of measurements present in the data set does not make this an easy task. The goals of this book are thus fourfold. The first is to create a useful reference on the applications of these statistical pattern recognition methods to the sciences. The majority of our discussions center around the fields of chemistry, geology, environmen­ tal sciences, physics, and the biological and medical sciences. In Chapter IV a section is devoted to each of these fields. Since the applications of pattern recognition tech­ niques are essentially unlimited, restricted only by the outer limitations of.

I Philosophical Considerations and Computer Packages.- I.1. Philosophical Considerations.- Data Considerations.- Pattern Recognition Approach.- Pattern Recognition Questions.- Pattern Recognition Nomenclature.- Variable Coding.- Categorization of Data.- Considerations to Keep in Mind.- Programs to Be Discussed.- Possible Approaches.- I.2. Biomedical Computer Program (BMDP).- Program Groups.- BMDP Control Language.- I.3. Statistical Package for the Social Sciences (SPSS).- Package Structure.- SPSS Programs.- SPSS Control Language.- I.4. ARTHUR.- ARTHUR Control Language.- ARTHUR Programs.- I.5. CLUSTAN.- CLUSTAN Format.- I.6. SAS.- SAS Programming.- Procedures Available.- II Pattern Recognition Approach to Data Analysis.- II.1. Preliminary Data Examination.- Step II.1a. Computer Compatibility.- Step II.1b. Eliminating Format Errors.- Step II.1c. Data Listing.- Step II.1d. Variable Distributions.- Step II.1e. Identifying Unique Cases.- Step II.1f. Variable Distributions After Elimination.- Step II.1g. Introductory Statistics.- Standard Scores.- II.2. Data Stratification.- t-Tests.- Step II.2a. Comparison of Two Groups.- Analysis of Variance.- Step II.2b. Comparisons of All Groups Simultaneously.- Multiple Way Analysis of Variance.- Step II.2c. Variable Subgroups.- CHI-Square Test.- Other Similar Programs.- II.3. Inter Variable Relationships.- Correlation Coefficients.- Step II.3a. Calculation of Correlation Coefficients.- Correlation Results.- Step II.3b. Bivariate Plots.- Partial Correlations.- Step II.3c. Partial Correlations.- Step II.3d. Clustering of Variables.- Results of the Clustering.- Canonical Correlations.- Step II.3e. Application of Canonical Correlations.- Regression Analysis.- Step II.3f. Regression Analysis.- Stepwise Regression.- Summary.- PLS-2.- Nonparametric Statistics.- II.4. Unsupervised Learning Techniques.- Cluster Analysis.- Data Set Considered.- Step II.4a. Minimal Spanning Tree.- Drawing of the Tree.- Cluster Definition.- Minimal Spanning Tree Results.- Step II.4b. Hierarchical Clustering.- Description of the Dendrogram.- Other Hierarchical Clustering Programs.- Step II.4c. Nonlinear Mapping.- Summary.- Step II.4d. CLUSTAN.- Hierarchical Techniques in CLUSTAN.- Other Clustering Techniques.- Summary.- II.5. Supervised Learning Techniques.- Step II.5a. k-Nearest Neighbors.- Testing Supervised Results.- Step II.5b. Discriminant Analysis.- Results from Discriminant Analysis.- Other Discriminant Programs.- Comparison of Discriminant Programs.- II.6. Variable Reduction.- Tools from Previous Steps.- Step II.6a. Selection of Variables.- Principal Component and Factor Analysis.- Step II.6b. Principal Component Analysis.- Classical Factor Analysis.- Data Considerations.- Factor Choices.- Step II.6c. Classical Factor Analysis.- Factor Rotations.- Underlying Variable Factor Analysis.- II.7. Data Manipulations.- SPSS Capabilities.- BMDP and ARTHUR Alterations.- III Implementation.- III.1. Typical SPSS Runs.- Card Deck.- Data Format Statement.- Missing Data.- Statistical Procedures.- Category Definitions.- File Saving.- III.2. Typical ARTHUR Runs.- Input Card.- Task Definition Cards.- Test Data.- III.3. Typical BMDP Runs.- Card Deck.- Variable Information.- Data Grouping.- III.4. Spss Implementations.- Cancorr.- Condescriptive.- Crosstabs.- Discriminant.- Factor.- Frequencies.- Nonpar CORR.- Partial CORR.- Pearson CORR.- Regression.- Scattergram.- T-Test.- III.5. ARTHUR Implementations.- Correl.- Distance.- Hier.- Karlov.- KNN.- Multi.- NLM.- Plane.- Scale.- Select.- Tree.- Varvar.- Weight.- III.6. BMDP Programs.- BMDP1D.- BMDP2D.- BMDP3D.- BMDP4D.- BMDP5D.- BMDP6D.- BMDP1M.- BMDP2M.- BMDP4M.- BMDP6M.- BMDP7M.- BMDP1R.- BMDP2R.- BMDP4R.- BMDP6R.- BMDP3S.- IV Natural Science Applications.- Biological Applications.- Medical Applications.- Geological and Earth Science Applications.- Environmental Applications.- Physics Applications.- Chemical Applications.- Summary.- References.- Appendix I. Pattern Recognition Definitions and Reference Books.- Appendix II. The Multivariate Normal Distribution.- Appendix III. Data Base Description.- Appendix IV. Indices of BMDP, SPSS, and ARTHUR Packages.- Appendix V. Programs, Manuals, and Reference Information.- Appendix VI. Summary of Analyses and Program Cross-Reference for Chapter II.- Appendix VII. Nonparametric Statistics.- Appendix VIII. Missing Values.- Appendix IX. Standard Scores And Weightings.- Computer Program Index.