Vertical Database Design for Scalable Data Mining

For several decades and especially with the preeminence of relational database systems, data is almost always formed into horizontal record structures and then processed vertically (vertical scans of files of horizontal records). This makes good sense when the requested result is a set of horizontal records. In knowledge discovery and data mining, however, researchers are typically interested in collective properties or predictions that can be expressed very briefly. Therefore, the approaches for scan-based processing of horizontal records are known to be inadequate for data mining in very large data repositories (Han & Kamber, 2001; Han, Pei, & Yin, 2000; Shafer, Agrawal, & Mehta, 1996).