Klaus Truemper
University of Texas at Dallas

Obtaining Explanations from Numerical Data

The explanation problem demands the construction of two explanations that tell why the numerical records of one population differ from those of a second population. The explanations must be simple enough that humans can comprehend them for decision making.

A new multi-step method called EXARP solves the explanation problem. The steps carry out discretization of data, selection of important factors, construction of explanations, and validation. While these terms are familiar in machine learning, the actual processes are quite different from prior ones. In particular, in each step a so-called alternative random process (ARP) is introduced that attempts to distort information or disrupt the computing process. Appropriate action by the solution algorithms prevent the ARPs from reaching these goals.

Applications abound in bioinformatics, economics, engineering, finance, and medicine. A number of tests in various areas have proved EXARP to be effective even when data sets are very small. The lecture focuses on some examples of medicine.