Sensitivity Analysis of MODENAR Method for Mining of Numeric Association Rules

Abstract

There are many association rules mining studies that focus on datasets consisting of only discrete or binary valued attributes. However, the data in many real-world applications are generally composed of quantitative or numeric values and classical association rule mining methods do not automatically work without preprocessing that disrupt the real data. It becomes a very hard problem to automatically mine high-quality rules in terms of many metrics. Recently, some researchers have considered numeric association rule mining as a multi-objective problem that best meets different criteria at the same time. In this paper, the parameter analysis of differential evolution based MODENAR, which is one of the few methods using multi-objective evolutionary algorithms for numeric association rules and aims to maximize support, confidence, comprehensibility, and amplitude of the ranges of attributes. For this purpose, the effects of the parameters of MODENAR such as population size, crossover rate, threshold solutions, and weight for support and confidence to the number of rules obtained, average support, confidence, lift, conviction, certainty factor, netconf, yulesQ, and the coverage percentage in five real-world data whose attributes consist of numeric values are carried out for the first time in this study.