Analysis of Various Interestingness Measures in Class Association Rule Mining

Abstract

Many measures have been developed to determine the interestingness of rules in data mining. Numerous studies have shown that the effects of different measures depend on the concrete problems, and different measures usually provide different and conflicting results. Therefore, selecting the appropriate measure becomes an important issue in data mining. In this paper, a novel approach to select the appropriate measure for class association rule mining is proposed. The proposed approach is applied to several problems, including benchmark and real-world datasets. The experimental results show that the proposed approach is a powerful tool to analyze various measures to select the right ones for the concrete problems, leading to the increase of the classification accuracy. Based on the study, this paper further proposes four properties of interestingness measures that should be considered in class association rule mining.

An extended probabilistic model building genetic network programming using both of good and bad individuals

“…Equation (1) ensures that both the information on the node connections and the reusability of the node connections are taken into account to construct the probabilistic model of PMBGNP. Previous work on PMBGNP has testified its superiority to solve some problems over conventional algorithms, such as data mining [8,13] and robot control [14].…”

“…PMBGNP inherits the advantages of EDAs owing to its probabilistic modeling, and the features of its graphbased representation ensure its higher expression ability than the conventional EDAs. It has been proven to successfully solve some real problems, such as data mining [12,13] and reinforcement learning (RL) problems [14]. As in the case of most of the other EDAs, PMBGNP uses truncation selection to only select the good individuals for probabilistic modeling; the bad individuals are just abandoned to be considered as useless.…”

Creating stock trading rules using graph-based estimation of distribution algorithm

“…PMBGNP applies a recent proposed EC technique, i.e., genetic network programming (GNP) [26], [27], [28], as the base model to construct its individuals. Following the research directions of EDA, the univariate [22], pairwise [29] and continuous PMBGNP [30] were proposed previously, which have also been studied in both theory [31] and applications [32], [33]. Different from the conventional EDA techniques, PMBGNP is dedicated to solve a different sort of problems, that is, the intelligent agent control.…”

Continuous probabilistic model building genetic network programming using reinforcement learning

“…2. Due to the unique features of its graph structures, PMBGNP explores the applicability of EDAs to wider range of problems, such as data mining [22,14,23] and the problems of controlling the agents' behavior (agent control problems) [16,[24][25][26].…”