The effect of threshold values on association rule based classification accuracy

Abstract

Classification Association Rule Mining (CARM) systems operate by applying an Association Rule Mining (ARM) method to obtain classification rules from a training set of previously classified data. The rules thus generated will be influenced by the choice of ARM parameters employed by the algorithm (typically support and confidence threshold values). In this paper we examine the effect that this choice has on the predictive accuracy of CARM methods. We show that the accuracy can almost always be improved by a suitable choice of parameters, and describe a hill-climbing method for finding the best parameter settings. We also demonstrate that the proposed hill-climbing method is most effective when coupled with a fast CARM algorithm such as the TFPC algorithm which is also described.

Classification Inductive Rule Learning with Negated Features

“…Rules with a Laplace estimation accuracy lower than the threshold of 50% were removed from the ruleset. The performance of our inductive rule learner, with each of the different proposed rule refinement strategies, was compared with the TFPC associative rule learner of Coenen and Leng [5]. The evaluation metric used is the average accuracy produced using ten fold cross validation.…”

“…A criticism of the ARL approach is that a great many rules are generated, typically many more than in the case of IRL systems such as the approach proposed in this paper. However, to evaluate the IRL approach proposed in this paper, comparisons are made with the ARL technique using the TFPC algorithm [5] because we are interested in interpretable rules as a classifier and also because our experiments here focus on multi-class classification instead of binary classification. Therefore, we will not compare with the support vector machine (SVM) [9] method (reported to be one of the best method for text classification).…”

“…The rule is interpreted as, "if a and b exist in a record, then classify the record as belonging to class k". Examples of rule-based classifiers include Associative Rule Learning (ARL) systems such as Classification based on Multiple class-Association Rules (CMAR) [12], Classification based on Predictive Association Rules (CPAR) [17], or Total From Partial Classification (TFPC) [5] and Inductive Rule Learning (IRL) systems such as First Order Inductive Learner (FOIL) [13] and Repeated Incremental Pruning to Produce Error Reduction (RIPPER) [6].…”

A Rule Induction Algorithm for Knowledge Discovery and Classification

“…In these algorithms, approaches such as divide and conquer, association rules, covering rule induction, decision trees, and naïve Bayes are used [1][2][3][4][5][6][7][8][9][10][11][12][13]. Apart from these, there are also other algorithms that generate rules using techniques such as neural networks, ant colony optimization, genetic algorithms, particle swarm optimization, fuzzy logic, and support vector machines [14][15][16][17][18][19].…”

Statistical Identification of Key Phrases for Text Classification

“…All experiments described in this paper were undertaken using the authors' TFPC algorithm [2] [3]. TFPC (Total From Partial Classification) is a CARM algorithm that constructs a classifier by identifying Classification Association Rules (CARs) from a set of previously-classified cases.…”