Please use this identifier to cite or link to this item:
Other Titles: Algorithmic support for parallel association rule search methods
Authors: Дмитриева, О.А.
Половинка, О.Л.
Keywords: алгоритм поиска
ассоциативное правило
обнаружение знаний
метод Apriori
параллельная реализация
технология MapReduce
Issue Date: Nov-2018
Publisher: Донецький національний технічний університет
Abstract: В статье рассмотрены современные методы поиска ассоциативных правил и обнаружения знаний в больших объемах данных. На основе проведенного сравнительного анализа и учета специфики предметной области обосновано использование алгоритма Apriori, для которого рассмотрены критерии и способы распараллеливания при реализации процедуры поиска ассоциативных правил. Для повышения эффективности работы алгоритма при генерировании частых одноэлементных наборов предложено использование технологии MapReduce.
Description: The article presents a comparative analysis of modern methods of searching for associative rules, taking into account the specifics of the subject area, since the correct choice of a method and algorithm for processing knowledge determines the quality and reliability of the information obtained. The approaches that are used to association rule search for Big data are investigated. For solving this problem using of the Apriori algorithm is substantiated. The main shortcomings of the Apriori algorithm are identified. The basic directions of effectiveness increasing are proposed, which are focused on parallelizing the processing of Big data. The proposed solution will significantly reduce the scanning time. The criteria for the effectiveness of parallel implementations of the association search algorithms are formulated. As the main approaches to the distribution of data processing in the association rules search, such methods as method of placing data in memory systems, method of parallelization (according to tasks, data, hybrid method), methods of distribution and load balancing (dynamic, static or hybrid balancing) are considered. Setting the spectrum of threshold values as in ballistic methods with further refinement (adjustment) for each indicator is proposed as another direction to increase the efficiency of the algorithm. A parallel implementation of the Apriori algorithm is proposed to be performed by MapReduce technology. This will allow to use parallel mechanism for generating frequent single-element kits to form an optimal procurement plan in pharmacy chains. As a subject area in the work is used a database of pharmaceutical products sold in pharmacy chains. For the formation of frequent single-element sets based on the available information on the pharmacy sales turnover, a transaction base is formed. To avoid shortages, it is suggested to introduce and automatically track the minimum amount of rarely ordered medicines.
ISSN: 2075-4272
Appears in Collections:Кафедра прикладної математики та інформатики

Files in This Item:
File Description SizeFormat 
025_dmitrieva.pdf1,11 MBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.