Association rule mining allows you to find a patterns between related events. The first algorithm for mining association rules, called AIS was developed in 1993 by the experts of the IBM Almaden Research Center. This pioneering work has increased the interest in association rules; the middle of the 90s of the last century was the peak of the research in this area, and since then every year there are several algorithms.
Use free sample research papers on Association rule mining to know that for the first time this problem was proposed to association rule mining to find patterns for typical purchases made at supermarkets, so sometimes it is also called market basket analysis.
Algorithms for association rule mining are designed to find all the possible rules, and the support and reliability of these rules should be above some apriori defined thresholds, which are called the minimum support (minsupport) and the minimum reliability (minconfidence).
The problem of finding association rules is divided into two subtasks:
- Finding all sets of items that meet the minsupport threshold. These sets of elements are called frequent.
- Generation of a set of rules of elements found in accordance to claim p. 1, with accuracy that meets the threshold minconfidence.
One of the first algorithms to effectively solve such class of problems was the APriori algorithm. Beside this algorithm, there also has recently been developed a number of other algorithms: DHP, Partition, DIC, and some others.
Values for the minsupport and the minconfidence are chosen in such a way so that to limit the number of rules found. If the support is of great importance, the algorithms will find the rules that are well known to the analysts or are so obvious that there is no point in having such an analysis. On the other hand, a low support leads to the generation of a great number of rules which, of course, require substantial computing resources. However, most of the interesting rules are at the low support threshold. However, too low threshold leads to the generation of statistically unreasonable rules.
Association rule mining is not a trivial task as it might seem at first glance. One of the problems is the algorithmic complexity in finding frequent itemsets elements, as with the number of elements in I (| I |) increases exponentially the number of potential sets.
When association rule mining, we assume that all the analyzed elements are homogeneous.
In the market basket analysis, there are goods that have exactly the same attributes, except for the name. But it will not be too hard to add the information about the transaction, in which commodity group includes goods and build a hierarchy of goods.
These rules are called the generalized association rules.
Introduction of additional information about the grouping of elements into a hierarchy will have the following advantages:
- This helps to establish association rules, not only between the individual elements, but also between different levels of the hierarchy (groups).
- Individual members may have little support, but on the whole group can meet the threshold minsupport.
Are you looking for professional research paper help on Association Rule Mining topics? Check this custom writing service to learn more!