An intelligent supplier evaluation model based on data-driven support vector regression in global supply chain
Graphical abstract
Introduction
In supply chain management, the supplier is a fundamental player. An appropriate supplier would have a positive and lasting effect on the competitiveness of the entire supply chain (Araz and Ozkarahan, 2007, Chen et al., 2006). Therefore, supplier evaluation is one of the most vital issues in successful supply chain management (Ho, Xu, & Dey, 2010). Supplier evaluation process is complicated and difficult due to the consideration of multiple criteria (Karsak & Dursun, 2015). It has received considerable attention and has been studied extensively for decades (Chai, Liu, & Ngai, 2013).
Most traditional supplier evaluation models (Chen et al., 2006, Ho et al., 2010, Azadi et al., 2015) require much expert experience, and thus are highly sensitive to the quantity and quality of experts. In today’s highly competitive global world, there is a boom in the number of suppliers (Deng, Hu, Deng, & Mahadevan, 2014). This has contributed to two conflicts in the supplier evaluation process. One is a growing need facing an insufficient number of experts. The other is the growing evaluation requirement, yet there is a lack of excellent experts. In other words, the quality and quantity of involved experts are difficult to guarantee during the present supplier evaluation process. There is an urgent need for a novel intelligent supplier evaluation model.
To evaluate numerous suppliers effectively, a regression model is introduced for supplier evaluation in this paper. This regression model represents the relationship between multiple criteria and the final evaluation result. But the regression relationship is high-dimensional and non-linear. A regression model is therefore difficult to construct.
Due to the high dimensionality of the feature space (Drucker, Burges, Kaufman, Smola, & Vapnik, 1997), SVR has the superior performance in high-dimension non-linear regression problems (Vapnik, 2013). The rapid development of SVR in statistical learning theory encourages researchers to actively apply SVR to various research fields (Wu, Ho, & Lee, 2004). Since there are many successful applications in SVR, it motivates this research in using SVR to construct the regression model for supplier evaluation. However there are two difficulties in building a traditional SVR model. On is the labelling of training data, which must be done manually. The other is the three parameters of the SVR model should be adjusted with prior knowledge. In this paper, a data-driven SVR is proposed for overcoming the two mentioned difficulties to alleviate the pressure on experts.
For the first difficulty, an integrated MCDM approach is used to assess the supplier and help to label them for training. Extensive MCDM approaches have been proposed to asses the suppliers with multiple criteria (Chai et al., 2013), such as analytic hierarchy process, data envelopment analysis (DEA), technique for order preference by similarity to an ideal solution (TOPSIS), and their integrations. These methods evaluate suppliers in different aspects. Each of these methods has its advantages and disadvantages. In particular, integrated MCDM is gaining a wider application owing to the combination of strength of different approaches (Büyüközkan and Çifçi, 2012, Yazdani, 2014, Zavadskas et al., 2016). Especially for global supply chain, there are various suppliers and evaluation requirements. The supplier can not be evaluated by only one method in one single aspect. If only DEA is used to evaluate supplier, the given supplier would be evaluated in productive efficiency. But when the given supplier only focuses on productive efficiency, it would ignore the environment issue. TOPSIS is introduced to overcome this problem. The ideal supplier is defined as environmentally friendly, and the given supplier can be evaluated by calculating the distance between the given supplier and the ideal supplier. In this paper, the assessments of each supplier are obtained by an integrated data-driven MCDM using DEA, TOPSIS and the Goal. Instead of the manual labels, the labels of each supplier are obtained by the integrated MCDM approach.
For the second difficulty, genetic programming is employed for parameter adjustment. There are three key parameters for SVR: the kernel function , the penalty parameter C and the tolerable deviation . The kernel function is usually selected from some classical kernels, and thus would not fit all regression problems (Diosan, Rogozan, & Pecuchet, 2007). In this paper, the kernel function is created by genetic programming. Therefore, the three parameters of SVR are all set by genetic programming without prior knowledge.
Consequently, an intelligent model is proposed for supplier evaluation to address the above issue based on previous work (Yijun, Jun, Zhuofu, Xin, & Weirong, 2017). The previous work builds a DEA-adaboost model for supplier selection simply. In this paper, the proposed intelligent model is improved in both DEA and adaboost. First, the labels of each supplier are obtained by an integrated MCDM instead of DEA. Then, the suppliers and their labels are used to train the SVR model. During the training process, three suitable parameters combination of SVR is evolved by genetic programming. Finally, the regression model for intelligent supplier evaluation is constructed by data-driven SVR. The contributions of the work are as follows:
- •
To evaluate various suppliers in the global supply chain, an intelligent model is proposed. Because the quality and quantity of involved experts are difficult to guarantee, the proposed intelligent model relies on data to alleviate the pressure on experts.
- •
Instead of labelling the suppliers for training regression model manually, an integrated MCDM is adopted, which integrates DEA, TOPSIS and GOAL. The labels of each supplier are obtained by analysing data in three aspects: the efficiency evaluation, the distance evaluation and the accomplishment evaluation.
- •
The three parameters of the SVR model are all obtained based on data without prior knowledge. The kernel function , the penalty parameter C and the tolerable deviation of SVR are optimized by an evolutionary algorithm. The kernel function is not selected from some classical kernels, but created by genetic programming.
This paper proceeds as follows. Section 2 provides related work from two aspects, hybrid supplier evaluation models and the algorithms for the parameters adjustment of SVR. Section 3 presents the regression model for supplier evaluation. Section 4 presents the proposed intelligent model based on data-driven SVR. Section 5 describes an case study of supplier evaluation and its results. Finally, Section 6 draws a conclusion.
Section snippets
Hybrid supplier evaluation models
As the most important role of supply chain management, supplier evaluation has received a great deal of attention from practitioners and researchers. Machine learning has the ability to learn some concepts in a manner similar to human beings (Dai & Wang, 2018). Therefore the combination of machine learning and MCDM is the most prevalent solution for addressing this problem(Ho et al., 2010).
Wu (2009) attempted to integrate DEA with neural network and decision tree, for assessing supplier
The support vector regression model for supplier evaluation
For supplier evaluation, simple linear regression is shown in Fig. 1(a), the final evaluations of each supplier are determined by one criterion, e.g. economic criterion. But nowadays complex supplier evaluation is a high-dimension non-linear problem, and the final evaluation is determined by multiple criteria as shown in Fig. 1(b). SVR has superior performance in high-dimensional non-liner problem because of constructing a feature space representation. So SVR is introduced to solve the complex
An intelligent supplier evaluation model based on data-driven SVR
In order to solve the two difficulties mentioned in the previous section, an intelligent supplier evaluation model is proposed based on data-driven SVR. For the first difficulty, the labels of training data are obtained by an integrated MCDM approach. For the second, three main parameters of SVR are set by a GP algorithm. Finally, the intelligent supplier evaluation model is built. The overall process of our proposed model is shown in Fig. 2.
Simulation
In order to assess the performance of the proposed intelligent model, this section presents findings from its implementation in a big company–ARCIC. ARCIC wishes to evaluate suppliers through multiple raw criteria. The list of criteria are provided by managers (Azadi et al., 2015). The sub-evaluations of each supplier in multiple criteria are obtained in Table 1. The data set relates to inputs, outputs, and goals of the outputs. The input dimension is divided into the following categories (
Conclusion and future research
Supplier evaluation is a great challenge in global supply chain management. An intelligent supplier evaluation model is proposed to overcome this challenge. The construction process of the proposed intelligent model is divided into two phases. In the first phase, the final evaluations of each supplier are obtained by an integrated MCDM approach, which consists of three aspects. The evaluations of three aspects are all obtained through analysing data. In the second phase, an SVR regression model
Acknowledgment
This work was supported by National Natural Science Foundation of China (61873353, 61672537, 61672539, 61772558, 61502055) and the Fundamental Research Funds for the Central Universities of Central South University (2018zzts173).
References (45)
- et al.
An improved support vector regression model for estimation of saturation pressure of crude oils
Fluid Phase Equilibria
(2015) - et al.
Supplier evaluation and management system for strategic sourcing based on a new multicriteria sorting procedure
International Journal of Production Economics
(2007) - et al.
A new fuzzy dea model for evaluation of efficiency and effectiveness of suppliers in sustainable supply chain management text
Computers & Operations Research
(2015) - et al.
A novel hybrid mcdm approach based on fuzzy dematel, fuzzy anp and fuzzy topsis to evaluate green suppliers
Expert Systems with Applications
(2012) - et al.
Application of decision-making techniques in supplier selection: A systematic review of literature
Expert Systems with Applications
(2013) Support vector regression based on optimal training subset and adaptive particle swarm optimization algorithm
Applied Soft Computing
(2013)- et al.
A fuzzy approach for supplier evaluation and selection in supply chain management
International Journal of Production Economics
(2006) - et al.
Data envelopment analysis: Prior to choosing a model
Omega
(2014) - et al.
Automatic magnetic resonance image segmentation by fuzzy intercluster hostility index based genetic algorithm: An application
Applied Soft Computing
(2016) - et al.
Supplier selection using ahp methodology extended by d numbers
Expert Systems with Applications
(2014)
Chaotic particle swarm optimization algorithm in a support vector regression electric load forecasting model
Energy Conversion and Management
Multi-criteria decision making approaches for supplier evaluation and selection: A literature review
European Journal of Operational Research
A hybrid group decision support system for supplier selection using analytic hierarchy process, fuzzy set theory and neural network
Journal of Computational Science
An integrated fuzzy mcdm approach for supplier evaluation and selection
Computers & Industrial Engineering
An enhanced dea Russell graph efficiency measure
European Journal of Operational Research
A hybrid intelligent fuzzy predictive model with simulation for supplier evaluation and selection
Expert Systems with Applications
Environmental and social criteria in supplier evaluation – Lessons from the fashion and apparel industry
Journal of Cleaner Production
Supplier selection: A hybrid model using dea, decision tree and neural network
Expert Systems with Applications
Type-2 fuzzy multi-objective dea model: An application to sustainable supplier evaluation
Applied Soft Computing
Real time face detection and facial expression recognition: Development and applications to human computer interaction
Computer Vision and Pattern Recognition Workshop
Handwritten numeral databases of indian scripts and multistage recognition of mixed numerals
IEEE Transactions on Pattern Analysis and Machine Intelligence
Hybrid support vector machine/hidden markov model approach for continuous speech recognition
Cited by (30)
Data-driven supplier evaluation in complex product industry value chain for brand value co-creation
2024, Computers and Industrial EngineeringArtificial intelligence techniques for enhancing supply chain resilience: A systematic literature review, holistic framework, and future research
2023, Computers and Industrial EngineeringAn integrated machine learning and MARCOS method for supplier evaluation and selection
2023, Decision Analytics JournalA barrier evaluation framework for forest carbon sink project implementation in China using an integrated BWM-IT2F-PROMETHEE II method
2023, Expert Systems with ApplicationsA fuzzy multi-objective decision-making model for global green supplier selection and order allocation under quantity discounts
2023, Expert Systems with ApplicationsA comprehensive framework and literature review of supplier selection under different purchasing strategies
2022, Computers and Industrial EngineeringCitation Excerpt :Talluri and Narasimhan (2003) used hybrid DEA and Kruskall-Wallis test for suppliers clusterization. Cheng et al. (2020) developed support vector regression (SVR) integrated with DEA-TOPSIS. The remaining approaches are found to be less explored.