Authors: Waleed Alsabhan
Addresses: University of Sharjah, P.O. Box 27272, Sharjah, UAE
Abstract: Association rule mining is applied to large databases to identify product associations. In the resulting large number of rules, interestingness is difficult to determine. Researchers have defined various measures of 'interestingness' such as support, confidence, lift and gain. Support is the probability of occurrence of an item or set of items, and is the most important of these measures, since the other measures are calculated using support. This current research suggests some deficiencies in the support measure and shows it is not consistent with its definition. Because other measures are calculated using support, this may make the other measures inconsistent. The researcher in this study proposes a new measure called normalised support, which is normalisation of general support, in other context-adjusted support or penalised support. Normalised support recommendations can stabilise product sale by product cross-sell promotion. In addition, the usefulness of other measures improves automatically.
Keywords: data mining; databases; interestingness measurement; product associations; rule mining; normalised support; market basket analysis; support; confidence; lift; gain; probability; item occurrence; item sets; support measures; inconsistent measures; normalisation; context-adjusted support; penalised support; support recommendations; product sales; cross-sell promotion; grocery stores; data analysis techniques; data analysis strategies.
International Journal of Data Analysis Techniques and Strategies, 2012 Vol.4 No.1, pp.101 - 114
Received: 08 May 2021
Accepted: 12 May 2021
Published online: 26 Jan 2012 *