A Hybrid Parallel Processing Strategy for Large-Scale DEA Computation

Using data envelopment analysis (DEA) with large-scale data poses a big challenge to applications due to its computing-intensive nature. So far, various strategies have been proposed in academia to accelerate the DEA computation, including DEA algorithms such as hierarchical decomposition (HD), DEA...

Full description

Saved in:
Bibliographic Details
Published in:Computational economics Vol. 63; no. 6; pp. 2325 - 2349
Main Authors: Chang, Shengqing, Ding, Jingjing, Feng, Chenpeng, Wang, Ruifeng
Format: Journal Article
Language:English
Published: New York Springer US 01.06.2024
Springer
Springer Nature B.V
Subjects:
Algorithms
Behavioral/Experimental Economics
Computation
Computer Appl. in Social and Behavioral Sciences
Computing time
Data analysis
Data envelopment analysis
Data processing
Density
Economic Theory/Quantitative Economics/Mathematical Methods
Economics
Economics and Finance
Math Applications in Computer Science
Message passing
Multiprocessing
Operations Research/Decision Theory
Parallel processing
Simulation
Time functions
Data envelopment analysis (DEA)
Parallel processing
Performance analysis
Hierarchical decomposition (HD)
Large-scale
Message passing interface (MPI)
ISSN:0927-7099, 1572-9974
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Using data envelopment analysis (DEA) with large-scale data poses a big challenge to applications due to its computing-intensive nature. So far, various strategies have been proposed in academia to accelerate the DEA computation, including DEA algorithms such as hierarchical decomposition (HD), DEA enhancements such as restricted basis entry (RBE) and LP accelerators such as hot starts. However, few studies have integrated these strategies and combined them with a parallel processing framework to solve large-scale DEA problems. In this paper, a hybrid parallel DEA algorithm (named PRHH algorithm) is proposed, including the RBE algorithm, hot starts, and HD algorithm based on Message Passing Interface (MPI). Furthermore, the attribute of the PRHH algorithm is analyzed, and formalized as a computing time function, to shed light on its time complexity. Finally, the performance of the algorithm is investigated in various simulation scenarios with datasets of different characteristics and compared with existing methods. The results show that the proposed algorithm reduces computing time in general, and boosts performance dramatically in scenarios with low density in particular.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0927-7099
1572-9974
DOI:10.1007/s10614-023-10407-1