M9190 Regression, classification and statistical learning

Faculty of Science
Autumn 2013
Extent and Intensity
2/0. 2 credit(s). Type of Completion: z (credit).
Teacher(s)
prof. Michael Schimek, Ph.D. (lecturer)
Guaranteed by
prof. RNDr. Ivanka Horová, CSc.
Department of Mathematics and Statistics – Departments – Faculty of Science
Contact Person: prof. RNDr. Ivanka Horová, CSc.
Supplier department: Department of Mathematics and Statistics – Departments – Faculty of Science
Course Enrolment Limitations
The course is offered to students of any study field.
Course objectives
Regression and classification methods based on the classical linear model and its extensions constitute a core part of multivariate statistics. Over the last two decades various new approaches have been developed which are more appropriate for those data we currently collect in business intelligence (BI), information technology (IT), and biotechnology applications. Most of them relax parametric model assumptions and add additional flexibility when our task is fitting models to quantitative observations. Others are destined to perform predictive tasks. The latter belong to the group of statistical learning procedures. A methodological and computational challenge is the huge size and the high complexity of many data sets of interest.
Modern regression and classification techniques heavily rely on efficient computing. Statistical learning blends concepts of statistics and machine learning. In the lectures selected statistical approaches and appropriate computer concepts are introduced. A deeper understanding can be gained in exercises applying the open source statistics and graphics environment R.
Syllabus
  • Typical applications of regression and classification techniques
  • Basics of decision theory
  • The concept of regression model fitting
  • Nearest-neighbour and other classification techniques
  • Quality of estimators
  • Subset selection in regression models
  • Shrinkage via ridge regression and the lasso
  • Complexity control, regularization, and penalization
  • The concept of statistical learning
  • Resampling with cross-validation
  • Resampling with the bootstrap
  • Basics of smoothing and connection to shrinkage
  • Generalized additive models as an important application of smoothing (when time allows)
Language of instruction
Czech
Further comments (probably available only in Czech)
Study Materials
The course is taught only once.
The course is taught: in blocks.
General note: Výuka bude probíhat v zasedací místnosti ÚMS v těchto termínech: 21.11. 13:00-15:00; 22.11. 10:00-12:00; 25.11. 14:00-16:00; 26.11. 10:00-12:00; 28.11. 10:00-12:00.
  • Enrolment Statistics (recent)
  • Permalink: https://is.muni.cz/course/sci/autumn2013/M9190