Empirical Evaluation of Efficient Periodic Pattern Mining Algorithms in Transactional Databases

  • Ruba Obiedat


Digital revolution along with the wide spread of computerized systems everywhere, leading to a massive increase in the amount of available data that needs to be analyzed, hence traditional and manual analysis becomes impossible. In the 1990s, “Data mining” terminology was introduced in order to analyze data and discover and extract knowledge or patterns from existing databases. Periodic Pattern mining is a data mining task interested in the discovery of patterns that periodically appear in database transactions. Traditional periodic pattern mining algorithms discover periodic patterns relying on a measure called “maximum periodicity” which presents the maximum time period or number of transactions between two consecutive occurrences of a pattern in a database. A pattern is considered as a periodic pattern if it doesn’t exceed the maximum periodicity threshold in all occurrences even once. This arise a big limitation and a very restricted condition which does not represent the real world situations as well. Consequently, several algorithms trigger this point and proposed a solution to relax this measure. This paper presents an experimental evaluation conducted on two recent algorithms supposed to solve this issue which are Periodic Frequent Pattern Miner (PFPM) and Stable Periodic-frequent Pattern-growth (SPP-growth). Four real-world datasets were chosen to examine the algorithms namely; Retail, Chess, Kosarak and Susy, two of them present a very huge dataset with a massive number of transactions. The two algorithms performance was compared for the first time in terms of execution time and memory consumption. The results showed that PFPM algorithm is more efficient in terms of execution time and memory utilization than SPP-growth algorithm for medium size datasets. The findings also showed that while the PFPM algorithm failed to handle very huge datasets, the SPP-growth algorithm performed better in those datasets. On other context, SPP-growth algorithm achieves better results in term of the number of found patterns than PFPM algorithm.

How to Cite
Ruba Obiedat. (2020). Empirical Evaluation of Efficient Periodic Pattern Mining Algorithms in Transactional Databases . International Journal of Advanced Science and Technology, 29(3), 12834 - 12853. Retrieved from https://sersc.org/journals/index.php/IJAST/article/view/30433