Expert Systems with Applications: An International Journal
Hi-index | 0.03 |
The variability of computer hardware is currently increasing, and there is a corresponding increase in the variability of statistical software. This is likely to continue in the foreseeable future, and indicates that there is an immediate need for a centralised collection of benchmark data sets. These sets should be sufficiently variable to enable a suitable choice for the flexible evaluation of software at a personal, or institutional, level. There are two major and interrelated reasons for having a collection of benchmark data, entailing a variety of available sets. These are for the quick and relatively easy evaluation of what the statistical package is actually doing, and for determination of numerical accuracy. Besides elaboration of these points, this articles gives a comprehensive review of the relevant and recent literature in which data sets have been proposed and/or used for evaluation purposes.