This course covers classical and modern theory and methodology for the analysis of multivariate data. Multivariate analysis ranges from simple visualization techniques to advanced classification and non-linear regression methods often found under the headings "machine learning", "datamining", or "pattern recognition". The techniques studied have applications anywhere multivariate data appear, including finance, bioinformatics and genomics, sociology, artificial intelligence, image processing, and many more.

The class will be organized to cover and relate topics in two areas:

1. "Classical" multivariate analysis - distribution theory, principal components, cluster analysis, discriminant analysis, factor analysis and canonical correlations.

2. "Modern" statistical learning and datamining - topics drawn from: Bayesian classifiers, nearest neighbor methods, local regression and adaptive splines, neural networks and projection pursuit, support vector machines and kernel methods, tree-based methods, cross validation, bagging, boosting, model averaging.

Course meetings: MW 9:10-10:30, 025 Old Chem Bldg.

Prerequisites: Stat 215 and 216; Stat 244 recommended; or consent of instructor.

Note: Permission numbers are required to enroll via ACES. These may be obtained by sending email to scs@stat.duke.edu or at the first lecture.