Learning Rule-Based Models - The Rough Set Approach
2014 (English)In: Comprehensive Biomedical Physics: Volume 6: Bioinformatics / [ed] Bengt Persson, Elsevier, 2014, Vol. 6, 19-39 p.Chapter in book (Refereed)
Rough sets are a mathematically well-founded approach to induce minimal rules from examples represented in the form of decision systems. Rough sets have been successfully used to build classifiers in many domains including lifer sciences. This chapter introduces the concept of rough sets in a semi-formal way accessible to a non-expert reader. Rough sets have several advantages over other approaches such as, for instance, human legibility of the rule models by non-experts, and the ability to represent combinatorial models. Other advantages of rough sets include data reduction and uncertainty handling. With help of examples taken from a very broad spectrum of bioinformatics applications of rough sets ranging from functional genomics, to proteomics, to transcriptomics, to epigenetics, to cancer and HIV research, this introduction explains the basics of rough sets. It then continues to more advanced topics in building rough set classifiers. The focus is on the process of developing a rule-based model, its interpretation, and statistical significance, which discern this chapter from many a text on machine learning. Finally, rough sets are briefly compared to other learning approaches including some statistical approaches. The availability of ROSETTA allows learning how to use rough set modeling in practice. A discussion of relative advantages and disadvantages of rough sets ends this introduction.
Place, publisher, year, edition, pages
Elsevier, 2014. Vol. 6, 19-39 p.
Cancer, Epigenetics, Feature selection and interaction, Genomics, HIV, Machine learning, Protein function
Computer and Information Science Cancer and Oncology
IdentifiersURN: urn:nbn:se:uu:diva-307101DOI: 10.1016/B978-0-444-53632-7.01102-3ScopusID: 2-s2.0-84943391090ISBN: 9780444536327 (print)ISBN: 9780444536334 (print)OAI: oai:DiVA.org:uu-307101DiVA: diva2:1048881