Frequent Pattern Mining : Overview

Frequent Pattern Mining or Association rule mining approach is commonly used for solving a wide range of data mining problems for obtaining interesting relations between objects and their features in big datasets. Association rule mining is mainly focuses on obtaining frequent patterns in transaction databases by applying Apriori-like algorithms Firstly, association rule mining was proposed as an effective tool for market basket analysis with the aim to analyze customer behavior and find typical patterns of purchases, which are made in supermarkets.

Thus, the main idea of this method is finding association rules, for instance,

                                            {𝑏𝑢𝑡𝑡𝑒𝑟,𝑏𝑟𝑒𝑎𝑑} ⇒ {𝑚𝑖𝑙𝑘},

what indicates the fact of buying butter and bread together will lead to buying of milk. These purchase patterns are used for organizing store space and product promotions with the aim to increase sales.

Formally, assume that 𝐼={𝑖1,…,𝑖𝑛} is a set of instances, 𝑇={𝑡1,...,𝑡𝑚) is a set of transactions, where each transaction 𝑡𝑖 is a subset of I (𝑇⊆𝐼), C is a subset I: 𝐶⊆𝐼. Association rule in the form of ⇒𝑌 , where 𝑋⊆𝐼, 𝑌⊆𝐼 and 𝑋 ⋂𝑌=∅.

Association rule is characterized by two parameters: support and confidence.

• Accordingly, support value of C with respect to T is defined as a number of transactions, which contains C:

                                         𝑠𝑢𝑝𝑝(𝐶) = {𝑡|𝑡∈𝑇, 𝐶∈𝑡 } (1)
  

• The confidence value of the rule 𝑋⇒𝑌 with respect to a set of transactions T, is the number of the transactions that contains X which also contains Y and it is define as:

                                   𝑐𝑜𝑛𝑓(𝑋⇒𝑌) = 𝑠𝑢𝑝𝑝(𝑋∩𝑌) / 𝑠𝑢𝑝𝑝(𝑋) (2)
  

_Support (1) and Confidence (2) _values are used as thresholds for selecting strong rules, which are greater than some minimum support and minimum confidence value.