Genetic algorithm with variable neighborhood search for the optimal allocation of goods in shop shelves

https://doi.org/10.1016/j.orl.2014.06.002Get rights and content

Abstract

Shelves on which products are being displayed are one of the most important resources in retail environment. The decision of shelf-space allocation and management is therefore a critical issue in retail operation management. In this paper a hybrid algorithm that combines a genetic algorithm with a variable neighborhood search is proposed to address the shop shelf allocation problem. Results obtained from an extensive experimental phase show the suitability of the proposed algorithm in addressing the problem at hand.

Introduction

The shelf-space allocation problem is a real-world problem faced by many retail companies. The problem arises when there is a large number of products to display, but with limited shelf space available at disposal. As reported in  [6], increasing sales by attracting the consumer’s attention and encouraging consumers to have additional purchase opportunities can be implemented by proper management of shelf-space allocation and product’s display. However, as reported in  [4], retail shelf-space management is one of the most difficult aspects of retailing due to several reasons. A significant reason is that while retail shelf space is fixed, the numbers of new potential products, customer wants, and competitors are constantly growing and evolving. At the same time, customers are consolidating shopping trips toward multi-purpose shopping  [7]. Thus, the success of any retailer depends on its ability to match its changing environment by continually deciding between how much of which products to shelve, where and when. Many retailers have now adopted software programs for creating allocation plans. However, as pointed out in  [13], because of the NP-hard complexity of the shelf-space allocation problem, many intuitive approaches have been applied in commercial systems for allocating shelf space within the retailing industry. The concern for simplicity and practicability of these commercial approaches, however, results in space allocation decisions that reach far from the optimal performance levels  [13]. Hence, considering the complexity of the problem, heuristic techniques seem to be the best road available to produce solutions to the problem in a reasonable computational time.

In this paper we propose a hybrid algorithm for the shelf-space allocation problem. The proposed algorithm combines a genetic algorithm with a variable neighborhood search. The use of a genetic algorithm, combined with a variable neighborhood search, results in an effective exploration/exploitation ratio for the considered problem, thus improving the quality of the final solutions.

The paper is organized as follows: Section  2 defines the problem addressed in this paper and presents its mathematical formulation. Section  3 provides a detailed description of the proposed method. Section  4 presents the experimental settings and the results, and finally Section  5 draws some final conclusions and suggests some hints for possible future work.

Section snippets

Problem definition

Before presenting the mathematical model considered in this paper, we introduce some definitions related to shelf-space allocation. The first term is a stock-keeping unit (SKU) which is used to uniquely identify a specific product or goods. The SKU is the smallest management unit in a retail store. Facing is a very important variable for shelf-space allocation. The number of the facings of an SKU is defined as the quantity of an item that can be directly seen on the shelves or fixtures by the

Method

In this section we describe the proposed shop shelf allocation algorithm. The algorithm is a hybrid algorithm that combines the genetic algorithms and a variable neighborhood search (VNS) algorithm. All the components of the algorithm are briefly described in the following sections and, finally, the resulting algorithm is outlined.

Experimental settings and results

In this section we compare the proposed shelf-space allocation algorithm against three other heuristic approaches described in  [4]. In the remaining part of the paper we will refer to these heuristics as H1, H2, and H3. The three algorithms are briefly described, but for all the details the reader is referred to  [4].

We generated the shelf space and product parameters consistent with the guidelines reported in  [8]. As done in  [4], we considered different problem sizes (i.e., different number

Conclusions and future work

In this work a hybrid genetic algorithm with a variable neighborhood search has been proposed for addressing the shelf-space allocation problem. The new algorithm has shown its suitability in producing optimal or sub-optimal allocation plans and its performance is better with respect to other existing heuristics defined to address this problem. The use of a genetic algorithm combined with a variable neighborhood search results in an effective exploration/exploitation ratio for the considered

References (13)

There are more references available in the full text version of this article.

Cited by (30)

  • A reduced variable neighborhood search-based hyperheuristic for the shelf space allocation problem

    2020, Computers and Industrial Engineering
    Citation Excerpt :

    In literature surveys, they are also known as nature-inspired metaheuristics (Du & Swamy, 2016), and some of them are classified as socio-inspired metaheuristics (Kumar & Kulkarni, 2019). In SSAP, some heuristics such as in Yang (2001) and Hansen, Raut, and Swami (2010), a greedy heuristic and genetic algorithm as in Urban (1998), simulated annealing as in Borin et al. (1994), and hybrid heuristics and metaheuristics as in Castelli and Vanneschi (2014) and Marshall, Johnston, and Zhang (2014) are used. The increasing attention to hyperheuristics has been partly motivated by some of the difficulties in using (meta)heuristics, but mostly in an attempt to automate the choice of suitable heuristics for the problem.

  • Shelf space allocation problem under carbon tax and emission trading policies

    2018, Journal of Cleaner Production
    Citation Excerpt :

    Extensions focusing on new models have been proposed with factors considered such as inventory (Hwang et al., 2005; Urban, 1998), locations and shelf level (Bai et al., 2013; Coskun, 2012), and joint optimization (Coskun, 2012; Katsifou et al., 2014; Murray et al., 2010). Moreover, several authors have proposed heuristics, metaheuristics, and hyper-heuristics for different shelf space problems (Bai et al., 2013, 2008; Hansen et al., 2010; Lim et al., 2004), large-scale assortment (Castelli and Vanneschi, 2014), and analysis (Irion et al., 2012; Leng et al., 2014). This present research uses the optimization model in Irion et al. (2012) as our base model.

  • Allocating products on shelves under merchandising rules: Multi-level product families with display directions

    2018, Omega (United Kingdom)
    Citation Excerpt :

    Lim et al. [13] further defended the linearity assumption stating that retailers prefer to operate on the linear portion of the S-shaped curve of marginal returns. This simplified version of the problem has been used by several authors to develop efficient heuristics and matheuristics for the problem [13,25–28]. Another key practical feature which has been neglected in the current state-of-the-art is the existence of product family arrangements on the shelves.

  • An innovative framework for designing genetic algorithm structures

    2017, Expert Systems with Applications
    Citation Excerpt :

    Generally, a GA consists of five main components, namely chromosome encoding, crossover, mutation, evaluation and selection. Since being introduced in 1975, there have been significant developments in the five components of GA: (1) chromosome encoding, e.g. binary encoding (He & Hui 2010), real encoding (Deep, Singh, Kansal, & Mohan, 2009), hybrid encoding (Chen, Xia, Liu, & Feng, 2006), variable length chromosome encoding (Dao & Marian 2011a), multi-dimensional chromosome encoding (Dao, Abhary, & Marian, 2014), etc.; (2) crossover, e.g. single-point crossover (Dao & Marian 2011b), multi-point crossover (Esen & Koç 2015), heuristic crossover (Balakrishnan, Cheng, Conway, & Lau, 2003), hybrid crossover (Suresh, Huang, & Kim, 2014), etc.; (3) mutation, e.g. adaptive directed mutation (Tang & Tseng 2013), power mutation (Deep & Thakur 2007), etc.; (4) evaluation, e.g. single objective evaluation (Faghihi, Reinschmidt, & Kang, 2014), multi-objective evaluation (Aiello, Scalia, & Enea, 2013), etc.; (5) selection, e.g. tournament selection (Castelli & Vanneschi 2014), roulette-wheel selection (Zhao & Wang 2011), dynamic ranking selection (Boudissa & Bounekhla 2012), etc. In fact, the components, i.e. chromosome encoding, crossover, mutation, evaluation and selection, have a certain impact on the efficiency and effectiveness of GA (Dao et al., 2016).

  • Formulation of customers' shopping path in shelf space planning: A simulation-optimization approach

    2016, Expert Systems with Applications
    Citation Excerpt :

    A heuristic method based on the problem-specific rules was applied to solve the fuzzy mixed integer non-linear goal programming model. Castelli and Vanneschi (2014) considering the number of facing proposed a hybrid algorithm to address the shop shelf allocation problem that combined GA with a variable neighborhood search. Chen, A-L. and Chang (2005) used data mining to identify customers’ behavior.

View all citing articles on Scopus
View full text