No Free Lunch in Data Fusion/Integration

Roland Soong
KMR, United States

Michelle de Montigny
KMR, United States


Given the proliferation of information and the near-impossibility of obtaining good single-source data, methods of data integration have assumed increased importance. A number of different methods of data integration have been proposed and even commercially realized. Clearly, there is a desire to form opinions about the accuracy of these methods in real-life applications.

It would be nice and clean if there is one single method that can be shown to be superior (or equal) to other methods. Unfortunately, at the heart of classification/learning theory is the famous No Free Lunch Theorem (see Duda, Hard and Stork 2001),1 which denies this intellectually lazy option (to wit, the free lunch).