Reliability and topology based network design using pattern mining guided genetic algorithm Nasim Nezamoddin, Sarah S. Lam ⇑ Department of Systems Science and Industrial Engineering, State University of New York at Binghamton, Binghamton, NY 13902, United States article info Article history: Available online 19 May 2015 Keywords: Genetic algorithm Network design Network reliability Pattern mining abstract This research proposes a new reliable network design methodology that is based on a pattern mining guided genetic algorithm (GA). The proposed method can be applied for a variety of applications includ- ing telecommunication, ad hoc, and power systems. In these networks, failures in certain parts of a net- work make it necessary for other parts to tolerate a higher traffic load in order to maintain adequate network connections. In addition, path changes due to dynamic routing of traffic can cause a time delay of communications in the network. To understand and reduce the connection failures costs, vigorous investigations are required to select the best design option under budget constraints. Given that many options for network topology and reliability allocation exist, a GA guided with pattern mining is proposed as an effective optimization method to design reliable network while considering link and node failures. Experimental designs under various assumptions have concluded that the guided GA approach is effec- tive in identifying a network solution within a short period of time. Ó 2015 Elsevier Ltd. All rights reserved. 1. Introduction The design of reliable networks is an important research area, particularly because of the serious problems in telecommunica- tions, ad hoc, and power system networks that result from connec- tion losses between the network sites. Connection failures in these networks often cause delays in communication, which degrade the overall system performance. Given that resources and budgets are limited, it is unrealistic to have an absolutely reliable network design. Therefore, there exists a tradeoff between network perfor- mance and the costs of designing and using a less reliable network. While network design varies according to the specific area of appli- cation, its basic concepts are consistent in all applications. An opti- mal design of network systems determines the best architecture with proper reliability indexes. Other variables including the net- work cost, load, reliability allocation limitations, and the inherent possibility of failure must be considered in the network design (Johnson, Lenstra, & Rinnooy Kan, 1978). A significant amount of research has been conducted to address the challenges in reliable network designs. Network reliability can be estimated using minimal cuts, Monte Carlo simulation, and artificial neural networks (Altiparmak, Dengiz, & Smith, 2009; Srivaree-ratana, Konak, & Smith, 2002; Yeh, Lin, & Yeh, 1994). One of the issues in network design problem is computation bur- den of testing the large number of possible designs. In general, net- work design problems are reported as NP-Hard problems (Kiu & McAllister, 1998). Meta-heuristics have received considerable attention and are considered as effective ways to search the design space and find the most reliable design (Dengiz, Altiparmak, & Belhin, 2010; Konak, 2012). For instance, a genetic algorithm (GA) -based reliable network system design combined with Monte Carlo simulation was studied in Altiparmak, Dengiz, and Smith (1998). GA was used for multiple-objective design optimiza- tion problems (Marseguerra, Zio, Podofillini, & Coit, 2005). Their research tried to determine the best design option when the com- ponents in the design could cause performance uncertainties. It was indicated that encoding strategies have the most significant effect on quality of the solution in meta-heuristic approaches (Chou, Premkumar, & Chao-Hsien, 2001). They tested the relation- ship between the effect of encoding, mutation, and crossover strategies on the performance of GA in finding the optimal solution for designing communication networks. GA was applied in variety of application specific network design problems including power systems (Leary, Srinivasan, Mehta, & Chatha, 2009), optical trans- port networks (Morais, Pavan, Pinto, & Requejo, 2011), and trans- portation systems (Gen, Kumar, & Kim, 2005). Due to GA’s flexibility and success in finding the best or near optimal in most of the network design problems, it has been considered as a main method for solving similar cases (Chen, Kim, Lee, & Kim, 2010). http://dx.doi.org/10.1016/j.eswa.2015.05.019 0957-4174/Ó 2015 Elsevier Ltd. All rights reserved. ⇑ Corresponding author. Tel.: +1 607 777 4754; fax: +1 607 777 4094. E-mail address: sarahlam@binghamton.edu (S.S. Lam). Expert Systems with Applications 42 (2015) 7483–7492 Contents lists available at ScienceDirect Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa