Correlation Associative Rule Induction Algorithm Using ACO

C Nalini

doi:10.4236/cs.2016.710244

Abstract

Classification and association rule mining are used to take decisions based on relationships between attributes and help decision makers to take correct decisions at right time. Associative classification first generates class based association rules and use that generate rule set which is used to predict the class label for unseen data. The large data sets may have many null-transac- tions. A null-transaction is a transaction that does not contain any of the itemsets being examined. It is important to consider the null invariance property when selecting appropriate interesting measures in the correlation analysis. Real time data set has mixed attributes. Analyze the mixed attribute data set is not easy. Hence, the proposed work uses cosine measure to avoid the influence of null transactions during rule generation. It employs mixed-kernel probability density function (PDF) to handle continuous attributes during data analysis. It has ably to handle both nominal and continuous attributes and generates mixed attribute rule set. To explore the search space efficiently it applies Ant Colony Optimization (ACO). The public data sets are used to analyze the performance of the algorithm. The results illustrate that the support-confidence framework with a correlation measure generates more accurate simple rule set and discover more interesting rules.

Highlights

Frequent pattern mining generates frequent patterns in a given dataset
The addition of a new kernel changes the shape of the mixed-kernel probability density function (PDF) curve and changes the values of the pheromones for the neighboring ranges
Data classification plays an important role in the data mining

Summary

Introduction

Frequent pattern mining generates frequent patterns in a given dataset. These patterns show interesting relationships among attribute-value pairs which occur frequently during frequent pattern mining. Classification Based on Multiple Association Rules (CMAR) [2] adopts FP-growth algorithm to generate a frequent pattern. CBA and CMAR adopt minimum support and confidence framework to implement frequent pattern mining Ant starts with an empty solution set and incrementally adds the terms one by one at a time based on the heuristic quality of the term (information gain) and the amount of pheromone deposited on it Baig et al proposed [11] a new ACO based classification algorithm called Ant Miner-C It implements a correlation-based heuristic function in order to guide the ant to select items that help to discover a rule. To improve the accuracy of the rule set and reduce the number of redundancies it applies correlation based heuristic function

The Proposed Algorithm

Result

Conclusion