Authors: Mohamed Azmi; Abdelaziz Berrado
Addresses: AMIPS Research Team, EMI, Mohammed V University of Rabat, B.P. 765, Rabat Agdal, Morocco ' AMIPS Research Team, EMI, Mohammed V University of Rabat, B.P. 765, Rabat Agdal, Morocco
Abstract: Classification based on association rules gets more and more interest in research and practice. In many contexts, rules are often mined from sparse data in high-dimensional spaces, which leads to large number of rules with considerable containment and overlap. Pruning is often used in search for an optimal subset of rules. This paper introduces a method for class association rules (CARs) pruning. It learns weights for a set of CARs by maximising the likelihood function subject to the sum of the absolute values of the weights. The pruning strength is controlled by a shrinkage parameter ⋋. The suggested method allows the user to choose the appropriate subset of CARs. This is achieved based on a trade-off between the accuracy and complexity of the resulting classifier which is controlled by changing ⋋. Experimental analysis shows that the introduced method allows to build more concise classifiers with comparable accuracy to other methods.
Keywords: class association rules; CARs; pruning; regularisation; Lasso; rules weighting; associative classification.
International Journal of Business Intelligence and Data Mining, 2021 Vol.18 No.2, pp.197 - 217
Received: 02 Mar 2018
Accepted: 28 Jun 2018
Published online: 28 Jan 2021 *