Towards the Patterns of Hard CSPs with Association Rule Mining

TL;DR

This paper applies association rule mining to analyze and identify patterns in the hardness of finite domain CSPs, a novel interdisciplinary approach combining constraint satisfaction and data mining techniques.

Contribution

It introduces a cascaded data mining approach to uncover interesting patterns in the hardness of randomly generated CSPs, a first in this research area.

Findings

Identified patterns correlating CSP characteristics with problem hardness

Demonstrated effectiveness of association rule mining in CSP analysis

Provided insights into the structure of hard CSP instances

Abstract

The hardness of finite domain Constraint Satisfaction Problems (CSPs) is a very important research area in Constraint Programming (CP) community. However, this problem has not yet attracted much attention from the researchers in the association rule mining community. As a popular data mining technique, association rule mining has an extremely wide application area and it has already been successfully applied to many interdisciplines. In this paper, we study the association rule mining techniques and propose a cascaded approach to extract the interesting patterns of the hard CSPs. As far as we know, this problem is investigated with the data mining techniques for the first time. Specifically, we generate the random CSPs and collect their characteristics by solving all the CSP instances, and then apply the data mining techniques on the data set and further to discover the interesting patterns of the hardness of the randomly generated CSPs