Rough Set Extensions for Feature Selection

摘要: Rough set theory (RST) was proposed as a mathematical tool to deal with the analysis of imprecise, uncertain or incomplete information knowledge. It is of fundamental importance artificial intelligence particularly in areas knowledge discovery, machine learning, decision support systems, and inductive reasoning. At heart RST idea only employing contained within data, thus unlike many other methods, probability distribution assignments are not required. relies on concept indiscernibility group equivalent elements generate granules. These granules then used build structure approximate given concept. This framework has unsurprisingly proven successful for application task feature selection. Feature selection (FS) term problem selecting input attributes which are most predictive outcome. Unlike dimensionality reduction algorithms preserve original semantics features following reduction. been applied tasks which involve datasets that contain huge numbers features (in order tens thousands), would be impossible to process otherwise. Recent examples such problems include text processing web content classification. FS techniques have also small medium-sized datasets discover information-rich features. The rough sets resulted efficient algorithms. However, due granularity approximations generated by approach there often resulting level uncertainty. uncertainty usually ignored (by nature very fact it `uncertain'). In this thesis, number methods attempt use improve performance extensions thereof FS. approaches two domain where reduction high importance; mammographic image complex systems monitoring. utility demonstrated compared empirically several dimensionality techniques. In experimental evaluation sections, shown equal classification accuracy when results obtained from unreduced data. Based new fuzzy-rough approaches, further developments presented thesis. first these nearest neighbour classifier real-valued data. technique evaluated imaging application. Also, novel unsupervised reduces eliminating those considered redundant. Both mentioned above, UFRFS employed monitoring