2 Akshat Surana, R. Uday Kiran and P. Krishna Reddy
databases. In a real-world database, some items reoccur frequently while others reoccur
relatively infrequent (or rarely). Furthermore, the rare items may have a larger reoc-
currence interval than the frequent items. If the items’ frequencies in a database vary
widely, usage of single minsup and single maxprd framework to discover periodic-
frequent patterns containing both frequent and rare items leads to “rare item problem”
[8] (discussed in Section 2). It has been shown in the literature that periodic-frequent
patterns consisting of rare items can provide useful information.
Example 1. In a supermarket, costly and/or durable goods such as soap and shampoo
are less frequently purchased than cheaper and/or perishable goods such as bread and
jam. However, soap and shampoo together can generate more revenue per unit than
bread and jam. Furthermore, the duration between two consecutive purchases of soap
and shampoo can be generally longer than the two consecutive purchases of bread and
jam.
A model based on multiple minsups and multiple maxprds framework has been pro-
posed in [5] to confront “rare item problem.” However, this model is computationally
expensive to implement because periodic-frequent patterns mined do not satisfy down-
ward closure property, i.e., not all non-empty subsets of a periodic-frequent pattern are
periodic-frequent.
In this paper, we have proposed an improved model to mine periodic-frequent pat-
terns with multiple minsups and multiple maxprds framework. A pattern growth ap-
proach has also been proposed for efficient mining of periodic-frequent patterns. The
periodic-frequent patterns mined with the proposed model satisfy downward closure
property. As a result the proposed model is computationally efficient than the model
discussed in [5]. Experimental results show that the proposed approach is efficient in
mining periodic-frequent patterns containing both frequent and rare items.
The rest of the paper is organized as follows. In Section 2, we discuss the back-
ground on mining periodic-frequent patterns in transactional databases. In Section 3, we
discuss the motivation and introduce the proposed model. A pattern-growth approach
to mine periodic-frequent patterns is discussed in Section 4. Experimental results are
reported in Section 5. The last section contains conclusions.
2 Background
2.1 Periodic-Frequent Pattern Model
Periodic-frequent patterns [7] are a class of user-interest based frequent patterns that
exist in a database. The basic model of periodic-frequent pattern mining is as follows.
Let I = {i
1
, i
2
, ··· , i
n
} be a set of items. A set of items X where X ⊆ I is called
a pattern (or an itemset). A transaction t =(tid, Y ) is a tuple, where tid represents a
transaction-id (or a timestamp) and Y is a pattern. A transactional database T over I
is a set of transactions, T = {t
1
, ··· , t
m
}, m = |T |, where |T | is the size of T in total
number of transactions. If X ⊆ Y , it is said that t contains X or X occurs in t and such
transaction-id is denoted as t
X
j
, j ∈ [1, m]. Let T
X
= {t
X
k
, ··· , t
X
l
} ⊆ T , where k ≤ l and
k, l ∈ [1, m] be the ordered set of transactions in which pattern X has occurred. Let t
X
j