Simultaneous Feature Selection and Tuple Selection for Efficient Classification

It is no longer news that data are increasing very rapidly day-by-day. Particularly with Internet becoming so prevalent everywhere, the sources of data have become numerous. Data are increasing in both ways: dimensions or features and instances or examples or tuples, not all the data are relevant though. While gathering the data on any particular aspect, usually one tends to gather as much information as will be required for various tasks. One may not explicitly have any particular task, for example classification, in mind. So, it behooves for a data mining expert to remove the noisy, irrelevant and redundant data before proceeding with classification because many traditional algorithms fail in the presence of such noisy and irrelevant data (Blum and Langley 1997). As an example, consider microarray gene expression data where there are thousands of features (or genes) and only 10s of tuples (or sample tests). For example, Leukemia cancer data (Alon, Barkai et al. 1999) has 7129 genes and 72 sample tests. It has been shown that even with very few genes one can achieve the same or even better prediction acABStrAct