Summary Nonlinear methods for accurate deviation detection (NOMAD)
Keywords Machine learning, research preparatory, deviation detection, unsupervised methods
TimeFrame The project should be done during the spring of 2014 and finished in late May. Ideally, the student(s) should start before the end of 2013.
References [[References::[1] Kriegel, Kröger, Zimek, "Outlier Detection Techniques", Tutorial, The 2010 SIAM International Conference on Data Mining,

[2] Sudjianto, Nair, Yuan, Zhang, Kern, Cela-Díaz, "Statistical Methods for Fighting Financial Crimes", Technometrics, vol 52, pp 5-19 (2010) [3] Schölkopf, Platt, Shawe-Taylor, Smola, Williamson, "Estimating the Support of a High-Dimensional Distribution", Neural Computation, vol 13, pp 1443–1471 (2001) [4] Guo, Chena, Tsai, "A boundary method for outlier detection based on support vector domain description", Pattern Recognition, vol 42, pp 77-83 (2009)]]

Prerequisites Learning systems, multivariate analysis, programming skills
Supervisor Thorsteinn Rögnvaldsson, Stefan Byttner
Level Master
Status Open

Generate PDF template

The setting is unsupervised detection of deviations in data. The initial research question is to explore the efficiency of a selected set of unsupervised deviation detection methods (of which at least one is a novel method). The project is software related. The student(s) must have experience with matlab, be skilled in mathematics and be able to work in an organized way. Deliverables: (a) A definition and collection of benchmark problems (b) A state-of-the-art summary of suitable methods (c) A set of results of selected methods on the benchmark problems (d) An analysis and conclusion from these results (e) A report The work is well suited for writing a short scientific paper in the end and submit it to a conference. The project is suitable for 1-2 persons with high work capacity and the ambition to show abilities for scientific or high-level development work. The field is huge so there is no problem to find enough individual work for 2 students.

Example work packages: (a) Define the problem and define what aspects that should be tested / explored. (b) Literature search for state-of-the-art methods (c) Implementation of selected methods for the study (d) Run experiments (e) Analyze and write report