Budapest University of Technology and Economics, Faculty of Electrical Engineering and Informatics

The course requires basic knowledge in data mining. (See also the course Data Mining: Models and Algorithms) Background in probability theory and linear algebra is important. Knowledge in combinatorics and algorithms is an advantage.

May be studied in the same semester as „Data mining - Part 1”

The aim of the course is to discuss advanced techniques of data mining with useful knowledge of related disciplines supporting real-world, especially bioinformatics data mining projects. By the end of the course, students will be able to analyze biological (genomic, microarray, pathway, protein, chemical) data sets using complex data mining methods.

1.  Advanced classification methods: Bagging, boosting, AdaBoost.
2.  Random forest. Implementation of models by WEKA.
3.  Support Vector Machine. Kernel methods, graph kernels. Protein function prediction.
4.  Similarity measures, fingerprint based similarity search. Sketches.
5.  Dimensionality reduction by spectral methods, singular value decomposition, low-rank                             approximation.
6.  Spectral clustering, bi-clustering for microarrays.
7.  Mixture models. Maximum likelihood estimators, EM-algorithm.
8.  Gauss Mixture Models. Midterm test.
9.  Search engines, web information retrieval, PageRank and beyond.
10. Rank learning for Protein Structure Prediction.
11. Text mining, natural language processing. Building databases and networks from PubMed and BioMed Central.
12. Graph mining algorithms. Frequent subgraph mining in microarray-based co-expression networks.
13. Semi-supervised classification of network data, graph stacking in biological networks.
14. Feature selection methods for unbalanced data sets. Final test.

Handouts, PowerPoint presentations, relevant research papers, web page, course mailing list and Wiki. Weekly regular office hour for consultations.

Case study: A practical problem. Choosing the model, solution method. Implementation of the algorithm.        

Grading principles:                    

Model:                               40% 

Solution method:                40%

May be studied in the same semester as „Data mining - Part 1”