Comparative Study of Various Sequential Pattern Mining Algorithms

Abstract

In Sequential pattern mining represents an important class of data mining problems with wide range of applications. It is one of the very challenging problems because it deals with the careful scanning of a combinatorially large number of possible subsequence patterns. Broadly sequential pattern ming algorithms can be classified into three types namely Apriori based approaches, Pattern growth algorithms and Early pruning algorithms. These algorithms have further classification and extensions. Detailed explanation of each algorithm along with its important features, pseudo code, advantages and disadvantages is given in the subsequent sections of the paper. At the end a comparative analysis of all the algorithms with their supporting features is given in the form of a table. This paper tries to enrich the knowledge and understanding of various approaches of sequential pattern mining.

A Comparative Study of Frequent Pattern Mining with Trajectory Data

“…In a sequence database, sequential patterns refer to itemsets arranged according to their inherent orders, such as time of occurrence [ 22 ]. An itemset is a subset of items, whereas an item is an entity that can have properties such as date, time, size, colour, speed, coordinates, etc.…”

“…Apriori exhibits a downward closure property, in which the key concept is that the support of a k-itemset is less than a threshold t if the support of all its subsets is less than that t. In other words, all subsets of any frequent itemset must also be frequent [ 22 ]. As a result, all of an itemset’s extensions will also be rare if the itemset itself is rare in the database.…”

“…The Apriori algorithm was proposed to deal with the problem of frequent itemset mining [ 23 ]. The algorithm is designed for mining frequently occurring itemsets by applying the downward closure property, which states that if an itemset is infrequent in the database, all its extensions would also be infrequent [ 22 ]. It is particularly useful during pruning by greatly reducing the search space.…”

“…The process continues until no further extensions can be made. This approach is also known as the level-wise or breadth-first approach [ 7 , 22 ]. Each n-sequence extension scans the whole database.…”

“…GSP adopts a similar type of horizontal database as Apriori and also uses the breadth-first search for frequent sequential pattern discovery. GSP attempts to generalize sequential pattern mining by employing time restrictions, sliding time windows, and taxonomies in sequential patterns [ 22 ]. Like AprioriAll, GSP follows a two-step approach where all candidates are generated prior to support counting.…”