Scalable classification over SQL databases
Surajit Chaudhuri, Usama M. Fayyad, Jeff Bernhardt · 1999
We identify data-intensive operations that are common to classifiers and develop a middleware that decomposes and schedules these operations efficiently using a backend SQL database. Our approach has the added advantage of not requiring any specialized physical data organization. We demonstrate the scalability characteristics of our enhanced client with experiments on Microsoft SQL Server 7.0 by varying data size, number of attributes and characteristics of decision trees.