1. Introduction
The sudden interruption in global supply chains due to the confinement during the COVID-19 pandemic undoubtedly impacted the profitability and permanence of several supplier companies, especially in emerging economies. Correspondingly, it has provided lessons in that decision-makers now know how to manage uncertainty and have learned how to be resilient to future disruptions and identify risks, improve their impacts, and quickly return to normal conditions after the interruption to sustain their functions; in addition, they have considered elementary operations in the performance of internal integration, continuity, agility, and even strategies for sustained competitive advantage [
1]. Moreover, a resilient, sustainable supply chain seeks and has the customer himself and his satisfaction as the ultimate verifier [
2].
The aspects that contribute to the resilience of the supply chain are multifactorial. However, sustainability has been considered one of them. In [
3], the authors paved the way by exploring and demonstrating sustainability as a factor of resilience in the supply chain, considering the environmental and social aspects. Based on a fuzzy multi-objective mathematical model, their analysis integrates resilience and sustainability into a strategic design to seek compensation solutions from a resiliently sustainable supply chain.
Focusing on the community dimension, Hooks, Macken-Walsh, McCarthy, and Power [
4] showed how the sustainable social dimension of social cohesion and cooperation among farmers provided supply chain resilience in Ireland’s rural farming sector. The same result appears in [
5]; by merging the criteria of the LEAN philosophy and the management of the green supply chain, the authors found that increasing efficiency by reducing resources in the supply chain favors resilience in the links of the food sector, thus bringing an impact of social welfare to the community with food supply and with sufficient control and stability in the local supplier companies of the base of the pyramid. Environmental and social sustainability favor the chain’s resilience, minimizing the scope of the so-called “domino effect” phenomenon. The above-mentioned describes the cascading effect triggered by the interruption of some link in the chain, gradually affecting the others. The work of Ivanov [
6] showed that taking care of social aspects favors the prompt recovery of efficiency, reduction in storage points, inventory security, management of social and environmental risk management, and the strengthening of factories and downstream providers.
Studies have already shown that collaboration with a socially sustainable partner in the production links is not just an aspect of social responsibility with an exogenous impact on company management. Instead, it has been shown that the culture and practices of sustainability in the supply chain are linked to internal strengthening and bring advantages to the operational performance of the supply chain [
7,
8]. Socially sustainable performance within the supply chain positively affects the company’s economic performance, as evidenced by a study in Sri Lanka, South Asia [
9]. Due to the convenience mentioned above, it has become an urgent issue for the supply chain to address social problems within the chain (upstream and downstream), especially concerning labor conditions in emerging countries such as health and safety, child and bonded labor, employee education, wages, human rights, gender diversity and discrimination, sanitation, philanthropy, engagement, community employment, as relevant social issues for the manufacturing sector [
8].
In the disruptive context of the pandemic, the social aspects of sustainability in the supply chain’s resilience have been notorious [
10]. In the context of the pandemic, the strategic rooting in strengthening the chains has been vital, from urgent preventive care for the health of society [
11] to regional support, where it has brought about a double advantage for resilience. Moreover, it sustained regional development, and in turn, the strengthening of the link with regional suppliers contributed to the recovery of the chain in emerging contexts in Brazil [
12].
Nevertheless, during the pandemic, the risk of internal society in companies has become more acute since it has had severe social implications for the livelihoods and well-being of workers and their families. In addition, due to the disruption in supply chains, audits for the care of social issues within companies decreased. The above is a vulnerable context for workers and employers seeking the survival of businesses at the expense of labor rights [
13]. Although the suggestion of the pre-pandemic and in-pandemic literature continues to invite the study of social sustainability in the supply chain and demonstrate the role that sustainability has within resilience, said correlation is not necessarily simple, linear, and synchronous. The consideration of resilience is usually with a short-term view, with an immediate reaction to incorporating efficiency. However, sustainable actions do not immediately affect the resilience and performance of the chain [
14], so it is necessary to study resilience within the chain from a complex perspective [
12]. One way of dealing with these confounding variables within the chain is the path taken by Soleimani et al. [
15], who propose a Genetic Algorithm to solve a fuzzy mathematical programming model for optimization purposes in the green supply chain. In a systematic review, we found that several studies have used fuzzy techniques to analyze sustainability and could be categorized into three groups: those from the business perspective within a sector, those specialized in social sustainability, and those that analyzed resilience within the supply chain. Among the studies in the business sector, those in the transport sector stand out, e.g., the research conducted by Bandeira et al. [
16], who evaluated sustainability in Brazilian urban transport, or that of Djekic et al. [
17], who proposed a sustainable index in the Serbian dairy transport sector. Other business studies have focused on social sustainability in the industrial sector, such as Rajak and Vinodh [
18] in India or Cao et al. [
19], who proposed techniques to assess the aspects of social sustainability in the manufacturing company. Furthermore, the works of Hendiani and Bagherpour [
20] and Rostamnezhad et al. [
21] evaluate the social aspects of sustainability in the construction sector. Several works on sustainability within supply chain resilience have shown how fuzzy models effectively handle data uncertainty in real scenarios. Some examples are the multi-objective fuzzy mathematical fuzzy model of Fahimnia and Jabbarzadeh [
3] in a case evaluated in a sportswear multinational based in Australia and China; or the mixed integral multi-objective nonlinear programming model used by Baghizadeh et al. [
22], in the Iranian tire industry and, recently, Lotfi et al. [
23], who used a hybrid fuzzy approach with stochastic programming in the study of the Sustainable Healthcare Supply Chain (
Table 1).
To the extent reviewed here, it is shown that the works that have used fuzzy techniques sustain a triple methodological constant:
Fuzzy mathematical programming that aims at operational optimization.
Fuzzy multi-criteria decision-making methods.
Fuzzy inference systems using expert knowledge or systematic mathematical tuning.
Our work seeks to differentiate itself by using fuzzy evolutionary knowledge, not to optimize but to take out-of-sample prediction of a bivariate predictive relationship between social sustainability and chain resilience based on a fuzzy engine based on machine learning.
Evolutionary Algorithms (EAs) are a powerful search technique in the field of artificial intelligence, where a fitness function guides a search through an evolutionary process. The fitness function maximizes or minimizes the desired objective function. Examples of approaches inside the EAs field are Genetic Programming (GP) [
30], Evolutionary Strategies (ES) [
31], and Genetic Algorithms (GAs) [
32]—the GAs being versatile for applications. The hybridization between Evolutionary Algorithms and Fuzzy Logic is known as Fuzzy-Evolutionary Systems [
33], Fuzzy-Genetic ones being widespread based on GAs. Therefore, Fuzzy Logic in hybridization with Evolutionary Algorithms to discover knowledge (fuzzy evolutionary approach) is convenient, which allows the elaboration of models to obtain crisp numerical values from the concept of fuzzy linguistic variables. Using a Fuzzy Inference System (FIS) requires expert knowledge to define their fuzzy knowledge base, a knowledge base that will have the biases of the expert who defines their knowledge. In the case of the absence of an expert, a systematic setup to minimize the FIS error can be performed. However, the approaches mentioned above are not guaranteed as optimal. We distinguish two optimal values in the optimization literature: local optima and global optima. Local optima are solutions that seem best in their near search region, while a global optimum is the best of all possibilities. Systematic configurations and expert knowledge lead to a local optima solution, leading to the not optimal performance of the FIS. Genetic Algorithms (GA) can escape local optima to improve even more, which is the principal motivation in this study—to use artificial intelligence (GAs) to automatically configure the knowledge base of the FIS without biases.
Two main areas of Fuzzy-Evolutionary systems appear in the literature: (i) the adaptation of parameters and operators for optimization [
34,
35], and (ii) the fuzzy control system design, using EAs to discover the knowledge base of a FIS. Our work is related to knowledge discovery, and we will mention some relevant publications next. In [
36], the gold price prediction is studied, the algorithm FURIA [
37] performs the fuzzy rules generation, and an evolutionary algorithm tunes their trapezoidal membership function parameters. The width of the membership functions of a controller for a robot trajectory is optimized in [
38]. Reference [
39] fine-tuned a Fuzzy Controlled Charging System for Lithium-Ion Batteries by a GA. The non-invasive measurement of a stroke volume was conducted through the discovery of fuzzy rules by a GA [
40]. Some other examples are crop plantation identification [
41], breaking systems [
42], and prediction of pipe failures in water supply networks [
43]. Furthermore, a considerable diversity of applications using Fuzzy-Evolutionary systems can be found in the works [
44,
45,
46,
47,
48,
49,
50,
51,
52,
53,
54], among others in the literature. In this work, we use a Genetic Algorithm to produce the set of fuzzy rules.
Several social sustainability works have dealt with real-life scenarios and show how fuzzy models deal effectively with data uncertainty. Some examples are the multi-objective fuzzy mathematical model of Fahimnia and Jabbarzadeh [
3], the multi-objective mixed-integer nonlinear programming model used by Baghizadeh, Pahl, and Hu [
22], and recently Lotfi, Kargar, Rajabzadeh, Hesabi, Özceylan [
23], who used a hybrid fuzzy with a stochastic programming approach, among others. In this work, as a methodology, the following steps are followed: (i) a survey of supply chain managers in Mexico and Chile for data collection, (ii) the collected data are unified and validated in second-order constructs through a Structural Equation Model by the Partial Least Squares method (SEM-PLS) with the SmartPLS software, (iii) two databases are constructed, one for Mexico and one for Chile, and integers represent the nominal values; (iv) a GA refines the fuzzy outputs of the FIS, substituting the need for expert knowledge using the Mexico database as training data; (v) finally, using Chile’s database as testing data, we compute the FIS error.
Figure 1 summarizes the methodology of the work.
Therefore, the objective of this paper is, through the application of a fuzzy evolutionary approach, to evaluate under uncertainty the predictive conditions of social sustainability factors in the resilience of the Latin supply chain in the context of the COVID-19 pandemic, with a representative case of Mexico and Chile. The study’s method, results, conclusions, and social and managerial drivers are presented.
4. Results
Using the statistical software SmartPLS 3.3.7 [
59], we performed a reliability analysis to assess the internal consistency of the latent variables. To better the consistency of the measures, we generated second-order coefficients for social sustainability and resilience measures in the supply chain. They result in a single construct of social sustainability. The measurement model of both the first and second models was validated.
At first, the individual reliability of the indicators was verified, where the expected threshold is an external load
; although, in social sciences, indicators with a load
are valid for reliability [
62]. In the first analysis of two indicators, the external loads obtained a value between 0.996 and 0.597. Then, we assessed the composite reliability of the construct, where there should be values > 0.70 [
62]. The values to evaluate the consistency of the composite reliability constructs ranged between 0.827 and 0.940 and, in terms of the Average Variance Extracted (AVE), the parameters were acceptable, ≥0.5. We check the discriminant validity of the constructs, ensuring that the constructs were empirically different from the other constructs included in the structural model. We use the heterotrait-monotrait (HTMT) correlation ratio, that is, the average of the correlations of the indicators between constructs, as well as within the same construct [
63]. The threshold value for the HTMT criterion was far from one with a limit of ≤0.800, thus concluding that there is discriminant validity [
62]. Finally, we evaluate the variance inflation factor (VIF), used to evaluate the collinearity of the indicators, where they had satisfactory values ≤ 3, according to [
62]. The model’s predictive power within the sample had an
of 0.318, while the values of
in the first-order model were irrelevant <0.02.
For the sake of parsimony in the explanation of the model and preparation for a mathematical explanation of the fuzzy knowledge logic technique, we decided to generate a construct of second molecular order from the measures related to social sustainability, prioritizing information on the common variance effect, that is, the variance shared by each of the dimensions [
64]. The coefficients of the latent variables were taken from the result of the validation of the first-order measurement model. The resulting values were validated.
Table 4 shows the values of satisfactory latent loads (
) for the case of the new construct of the second-order of social sustainability (SS), with satisfactory values that oscillate between 0.789 and 0.873. The latent variable of resilience in the supply chain (RES_SC), with a load value (
) of 1, is due to the one-dimensional condition of the latent variable coefficient. The construct items, coefficients, and latent variables have satisfactory composite reliability and AVE values. It is clear how the model within the sample substantially improves the effect size of
with a value of 0.438, going from an almost null effect to a medium one. The HTMT criterion also improved, being further from one, with a value of 0.591.
It is essential to notice that Genetic Algorithms are stochastic by nature, but we are not determining the performance of the GA; instead, we are benchmarking the performance of the FIS. FISs are deterministic; therefore, we can compute the exact relative error value produced by the best-found fuzzy knowledge database in the training phase against the test data (Mexico’s data). The best fuzzy rules found by the GA in the training phase are in
Table 5. With the rules from
Table 5, we compute an absolute error of 12.97, equivalent to about 13 examples wrongly classified from 81, with an 84% classification accuracy; however, this absolute error could be the sum of slight differences between crisp values, and for some decisions makers, these differences would be insignificant. Therefore, we have a predictive power of the resilience in the supply chains of more than 80%, based on features related to social sustainability. Our proposed FIS is a powerful tool for decisions and policymakers in organizations.
From the experimental results, we can assert that we have a second-order model with valid latent variable coefficients for each construct. The above allows parsimony in the data that are appropriate for generating the predictive rules of the fuzzy analysis. Nevertheless, the evaluated supplier cases are low for fair wages, production ethics, and employee training.
5. Discussion and Conclusions
Our fuzzy technique results showed the relevance of social aspects within the business sectors and were consistent with fuzzy evaluation works performed in the business sectors, as is the case of Rajak and Vinodh [
18] and Hendiani and Bagherpour [
20]. They found social aspects relevant to the company both inside and outside the company. Regarding the relevant aspects in the care of the internal society or human capital, issues such as health and safety practices resulted. Likewise, equal opportunities and education in the workplace also arose. In the aspects referring to the care of the external society or community, issues of support for safety, employment and health systems resulted. Support for smaller suppliers was also relevant.
In the studies focused on the supply chain, our results also show consistency with other findings, such as those of the work of Yıldızbaşı et al. [
28] in the automotive sector in Turkey, where the criteria of safety and occupational health practices, non-discrimination and social responsibility with the community and human rights were significant. Likewise, the research of Đurić et al. [
27], who employed a fuzzy AHP technique and from the risk perspective, showed the relevance of social aspects in the chain in terms of the risk to the resilience of unhealthy working environments and danger to the health and safety of employees, as well as issues such as human rights, inclusion, benefit to the communities surrounding the links in the chain and the focal companies. Specifically, in the context of the health crisis caused by the COVID-19 disease, there have been recent diffuse studies that evaluate the resilience of the supply chain, some focused on the environmental dimension of sustainability, highlighting aspects such as the optimization of natural resources in the chain (such as the work of Lotfi, Kargar et al. [
23]).
Recently, research on chain resilience has reinforced the social aspects of sustainability, such as that of Wang et al. [
65]. From a fuzzy logic based on experts, they determined that health and safety are relevant attributes of a socially sustainable supplier within the Vietnamese logistics sector. Likewise, the research of Nayeri et al. [
29] assesses, as a socially sustainable factor for chain resilience, the opportunity for permanent jobs and human rights security in the work area. Our results focused exclusively on the relevance of social issues of sustainability. We utilize a fuzzy technique that has scarcely been used in sustainable approaches and supply chain resilience. We propose a Genetic Algorithm for generating fuzzy evolutionary knowledge to predict the bivariate relationship between social sustainability and Latin supply chain resilience through artificial machine intelligence. Although the work of Soleimani et al. [
15] also proposes a Genetic Algorithm, they use it for fuzzy mathematical programming, and the distinction between their work and ours is that their model aims at optimization and focuses on the environmental aspect of sustainability and, our model, has an approach focused on prediction and the social aspect of sustainability. Within the sample that our systematic review of the literature yielded, we did not find a work that contains the criteria of our research. This research aimed to analyze the predictive power of social sustainability criteria on the performance of supply chain resilience in the context of the COVID-19 pandemic. In summary, based on fuzzy evolutionary learning, we conclude that the aspects of labor rights, health and safety, inclusion, and social responsibility in the Latin supply chain contribute significantly to the predictability of the supply chain resilience in the representative cases of Mexico and Chile. In terms of managerial implications, a call is made to decision-makers to resize social welfare issues within the links of the chain. These results demonstrate that caring for the welfare of people, inside and outside the supplier companies, is not an issue exogenous to the company. It is not mere external altruism but rather endogenous strategic aspects that impact the capacity to sustain the Latin supply chain in the face of current and future disruptions. We invite supply chain managers to strengthen the supply chain’s resilience by taking care of the indicators related to adequate working conditions, strict surveillance of labor rights violations, occupational health, and safety policies for the external society, such as the development of local suppliers, as well as philanthropic activities/selfless help and carrying out programs to generate skill development opportunities for unemployed youth. For future lines of research, we will continue confirming the predictive power of fuzzy evolutionary knowledge concerning the influence of social aspects of sustainability in the Latin supply chain in different contexts, even at different times to know if the results are consistent outside of the pandemic context.