Integrating data mining and fuzzy decision-making techniques for analyzing the key minimizing factors of carbon emissions

2.1 Literature review on carbon emissions

A significant majority of the studies in the literature have focused on economic and financial factors as the determinant of carbon emissions. There are many economic and financial factors that affect carbon emissions positively and negatively. The first of these factors is GDP and the changes that occur in GDP. Gross domestic product is one of the factors that have a reducing effect on carbon emissions. Shahbaz et al. [14] stated that the relationship between GDP and growth rates and carbon emissions differs in the short and long run. Realized economic growth mostly creates a positive impact on the environment in the long run, while it causes environmental degradation in the short run. Nguyen et al. [15] claimed that in this process, as the GDP and development levels of the countries increase, the impact of economic growth on the environment changes positively.

Some of the researchers underlined that there is a positive correlation between the economic size of the country and carbon emissions. According to the ARDL analysis put forth by Lan et al. [16], an increase in the economic size will significantly increase China’s carbon dioxide emissions in the long run. The short-term results of the presented study, on the other hand, show the opposite of the long-term results. On the contrary, Yu et al. [17] made a study with STIRPART model on green industrial parks in China and concluded that economic growth is the critical driver for increasing CO2 emissions. Similarly, Mujtaba et al. [18] concluded that there is a positive association between economic growth and carbon dioxide emissions.

Financial development and trade openness are the other economic factors that determine carbon emissions. Empirical findings indicate that increases in trade openness have a positive impact on environmental quality [19]. Similarly, Nguyen et al. [15] concluded that there is a strong relationship between trade openness and carbon emissions. According to the study conducted by Rjoub et al. [20], investigating the regulatory role of financial developments in carbon emissions, it has been revealed that financial developments are an important regulatory power on carbon emissions. On the contrary, according to the empirical findings obtained by Habiba et al. [21], which examines the effects of financial development on carbon emissions, it is concluded that financial development increases carbon emissions.

Another factor affecting carbon emissions is the building sector and urbanization, which is closely related to construction. It is very valuable that the construction sector, which cannot be reduced below a certain economic size due to the increase in population growth and urbanization, reduces carbon emissions. The use of energy-saving new technology products at all stages of the building sector’s life cycle means a reduction in total carbon emissions [22]. Yíldírím and Yíldírím [23] showed that the increase in construction sector activities plays a positive role in carbon emissions. Németh-Durkó [24] identified that urbanization has a positive effect on carbon emissions, meaning that the number of cities increases emissions. Similarly, Shah et al. [25] concluded in their analysis that urbanization increases carbon dioxide emissions significantly. On the other hand, Adebayo and Beton Kalmaz [26] did not find any evidence of a significant link between urbanization and carbon emissions in their study.

According to different scholars, technological innovations should also be taken into consideration in minimizing carbon emissions. Khan et al. [27] suggested that the combined effect of technological innovations and environmental taxes can play an important role in reducing carbon emissions. Jiang et al. [28] also examined the impact of international trade on carbon emissions and emphasized that the adoption of environmentally friendly technologies can be effective in reducing carbon emissions caused by international trade. Similarly, Adebayo et al. [29] examined the relationship between renewable energy, technological innovation, and economic growth in Brazil, revealing that environmentally friendly technologies are important in reducing environmental degradation. Chu [30], on the other hand, evaluated the determinants of ecological footprint to explore the impact of environmental technologies on ecological footprint in OECD countries and states that environmental technologies are an important tool in reducing environmental degradation. Likewise, Anwar et al. [31] used the STIRPAT model in their carbon emission study for E-7 countries and stated that technological innovations are effective in reducing environmental degradation. In other words, technological innovations improve environmental quality.

Moreover, carbon emissions can also be affected by nature. Sarwar et al. [32] have reached the conclusion that the results of forest investments have a negative and statistically significant effect on carbon emissions, in the model they created using two different proxies, forest investments and forest areas. This implies that higher forest investment is helping to control carbon emissions in China. On the contrary, forest area has reported a positive and significant relationship with carbon emissions, illustrating that an increase in forest area leads to an increase in carbon emissions. On the other hand, Li et al. [33] concluded in their study that the extraction of natural resources increases carbon emissions. Implementation of environmentally friendly logistics operations provides reduction of carbon emission stock [34]. In addition, Feng et al. [35] revealed in their study that the ocean has great potential to reduce carbon emissions. According to the study, the development of marine energy and low-carbon marine shipping may have more potential for emission reduction in China.

Both incentive and restriction policies to reduce carbon emissions are very valuable. These policies, which can be implemented through incentives, can be implemented through a range of activities ranging from environmental campaigns in government departments, colleges, schools and universities, television programs to the Internet [36]. Similarly, Lu et al. [37] determined that government can reduce carbon emissions by encouraging low-carbon technology innovations by using spatial autocorrelation method. Additionally, since trade openness increases carbon emissions and negatively affects the environment, governments need to draw legal frameworks in various economic areas. Governments should impose penalties on industries that negatively impact the environment and create an innovative and environmentally friendly legal framework for attracting FDI into the country [25].

The use of renewable energy sources is extremely important in reducing the carbon emission problem. Carbon emissions occur because of the release of greenhouse gases released during the combustion of fossil fuels into the atmosphere [38]. These gases remain in the atmosphere due to the greenhouse effect and cause climate changes throughout the world [39]. Renewable energy sources, on the other hand, are obtained from natural sources, unlike fossil fuels. Due to these mentioned points, these types of energy do not emit harmful gases to the atmosphere [40]. Therefore, the use of renewable energy sources plays an extremely important role in reducing carbon emissions and tackling climate change [41]. Dinçer et al. [42] made an evaluation for the renewable energy projects by constructing a novel decision-making methodology. They reached a conclusion that renewable energy investments should be increased to overcome carbon emissions problem. Moreover, Liu et al. [43] and Dinçer and Yüksel [44] also reached similar results in their studies.

2.2 Literature review on fuzzy decision-making models

Many different scholars preferred to use fuzzy multi-criteria decision-making techniques to make evaluation related to carbon emissions [45, 46]. Li et al. [47] focused on the factors regarding the green finance to reach carbon neutrality targets in China. For this purpose, fuzzy analytic network process methodology is taken into consideration to weight the criteria. Similarly, Ding et al. [48] created a novel decision-making model by using fuzzy grey methodology to minimize carbon emission problem. Moreover, Derse [49] evaluated carbon emission model with the help of DEMATEL technique. Dinçer et al. [50] focused on the carbon emission problem in E7 countries. In this framework, they integrated DEMATEL and TOPSIS techniques to overcome this problem. Similar to this study, many different decision-making techniques are used together in various studies for this purpose, such as CRITIC and DEMATEL [51], AHP and DEMATEL [52], DEMATEL and COPRAS [53] and VIKOR and DEMATEL [54]. On the other side, in some studies, different fuzzy sets are integrated to these techniques [55, 56].

2.3 Literature review results

After evaluating these studies, it is possible to reach the following conclusions.

The problem of carbon emissions has become very popular especially in the last years.

By considering these results, in this study, it is aimed to make a priority analysis for the determinants of carbon emission problem with a novel fuzzy decision-making methodology. With the help of this situation, more significant factors of this problem can be identified so that it can be possible to take specific actions to overcome this problem without having extra costs.

3.1 N-gram

The text mining part of the study consists of three stages: (1) data collection, (2) data preprocessing and (3) N-Gram tokenization. Firstly, the data set used in this study consists of abstracts of studies published in the ScienceDirect index with “carbon emission” in the title. Studies with carbon emission in the title are filtered by choosing 2019-2023 according to the year and research articles as the article type. As a result of filtering, 1711 articles are used for analysis in the study.

With respect to the preprocessing data, the abstracts data has gone through some preprocessing stages to be ready for use in the N-Gram Tokenization stage. The steps performed in the data preprocessing process are as follows [57]:

All uppercase letters in the text have been converted to lowercase.

Finally, three types of n-gram tokens, unigram, bigram, and trigram, are used on the data ready for tokenization because of preprocessing. Within the scope of this stage, the most repeated single, double, and triple words are found, and the number of repetitions was counted [59].

3.2 Spherical fuzzy CRITIC

The CRITIC method is used as a weighting method in multi-criteria decision making. This method calculates the weight of the criteria, considering the correlation between the criteria and the measure of variability in the criteria [10]. In this respect, it is possible to mention some advantages of the method. Primarily, it is considered as an objective weighting method because it calculates the weights over the “(alternatives)x(criteria)” matrix [11]. Therefore, it is stated that more realistic results can be achieved. Second, since it uses the linear correlation coefficient, it considers the interdependence of the criteria. Finally, it normalizes the decision matrix by considering the ideal values of cost and benefit criteria at the same time [12]. This also contributes to more consistent results. In this study, CRITIC technique is used together with Spherical fuzzy numbers. In this method, firstly, expert opinions are obtained. These evaluations are converted into fuzzy numbers by considering the linguistic expressions in Table 1. In this table, μ, v, and π refer to the membership, non-membership, and hesitancy degrees.

In the second stage, the decision matrix (D) is formed by taking the arithmetic average (SWAM) of the obtained Spherical fuzzy expert opinions with the help of Equations (2) [13]. In this context, D_si gives information about fuzzy numbers of the expert evaluations.

SWAM ( D s 1 , D s 2 , … , D Sn ) = { [ 1 - ∏ i = 1 n ( 1 - μ D si 2 ) 1 n ] 1 2 , ∏ i = 1 n v D si 1 n , [ ∏ i = 1 n ( 1 - μ D si 2 ) 1 / n - ∏ i = 1 n ( 1 - μ D si 2 - π D si 2 ) 1 / n ] 1 / 2 }(1)

D = [ ( μ 11 , v 11 π 11 ) ⋯ ( μ m 1 , v m 1 , π m 1 ) ⋮ ⋱ ⋮ ( μ 1 n , v 1 n , π 1 n ) ⋯ ( μ mn , v mn , π 1 mn ) ](2)

In the third step, the decision matrix is normalized with Equations (4), and as a result, the normalized matrix (X) is obtained. In these equations, V represents the maximum operator [11]. These equations are generated for positive and negative attributes respectively. In this process, (μ_, v_, π_) refer to the minimum optimal values whereas (μ+, v+, π+) indicate maximum optimal values.

x ˜ ij = 1 - μ ij 2 X μ - 2 + v ij 2 X v - 2 + π ij 2 X π - 2 μ ij 2 V μ - 2 + v ij 2 Vv - 2 + π ij 2 V π - 2 1 - μ - 2 X μ + 2 + v - 2 X v + 2 + π - 2 X π + 2 μ - 2 V μ + 2 + v - 2 Vv + 2 + π - 2 V π + 2(3)

x ˜ ij = 1 - μ ij 2 X μ + 2 + v ij 2 X v + 2 + π ij 2 X π + 2 μ ij 2 V μ + 2 + v ij 2 Vv + 2 + π ij 2 V π + 2 1 - μ - 2 X μ + 2 + v - 2 X v + 2 + π - 2 X π + 2 μ - 2 V μ + 2 + v - 2 Vv + 2 + π - 2 V π + 2(4) In the fourth stage, the correlation coefficient is calculated with the help of Equation (5) over the normalized matrix [11].

ρ ˜ jk = ∑ i = 1 m ( x ˜ ij - x ¯ ˜ j ) ( x ˜ ik - x ¯ ˜ k ) / ∑ i = 1 m ( x ˜ ij - x ¯ ˜ j ) 2 ∑ i = 1 m ( x ˜ ik - x ¯ ˜ k ) 2(5) In this scope, x ¯ ˜ j and x ¯ ˜ k explain the fuzzy average of j and k criteria. This average is computed by using Equation (6).

x ¯ ˜ j = 1 n ∑ j = 1 n x ˜ ij i = 1 , 2 , … , m(6)

The fifth stage includes the calculation of the standard deviation ( σ ˜ ) of each fuzzy number by Equation (7).

σ ˜ j = 1 n - 1 ∑ j = 1 n ( x ˜ ij - x ¯ ˜ j ) 2 i = 1 , 2 , … , m(7) In the sixth stage, the index (C) value is calculated with Equation (8).

C ˜ j = σ ˜ j ∑ k = 1 n ( 1 - ρ ˜ jk ) j = 1 , 2 , … , n(8) In the last stage, the weight (w) of each criterion is obtained by Equation (9).

w ˜ j = C ˜ j ∑ j = 1 n C ˜ j(9)

There are lots of different multi-criteria decision-making models to rank alternatives, such as VIKOR, TOPSIS and ARAS. However, these models are criticized by scholars in many ways. For example, the TOPSIS method uses Euclidean distance to get the distance to the optimal values. On the other hand, this technique has been criticized because it is not very accurate to consider the Euclidean distance when calculating the distance to the negative ideal [5, 6]. Also, there are many studies expressing the general problems of Euclidean calculus in the TOPSIS technique. In this context, studies that consider alternative distance calculation in this process are also included in the literature [7–9]. Similarly, the ARAS method uses the ratio of the sum of the weighted normalized values to the optimal sum after normalization. In this framework, since the ratio of similarities is considered, it is thought that using the geometric mean in proportional concepts will produce more accurate results.

In summary, there are some disadvantages of decision-making models used for alternative ranking in literature. It is seen that a new technique is needed to minimize these problems. In this framework, a new multi-criteria decision-making model is developed that can be considered for alternative rankings in this study. The steps of the model named Ranking Technique by Geometric Mean of Similarity Ratio to Optimal Solution (RATGOS) are given below.

Step 1: The decision matrix (D) is created. In this context, A represents m alternatives and C represents n criteria. The details are demonstrated in Equation (10).

D = C 1 … C n A 1 ⋮ A m [ d 11 ⋯ d 1 n ⋮ ⋱ ⋮ d m 1 ⋯ d mn ](10)

Step 2: For each criterion, the optimal value is determined by Equations (12).

Optimal = max i d for benefit(11)

Optimal = max i 1 d for cost(12)

Step 3: Each criterion is divided by the optimal value with the help of Equations (14). In this way, both normalization is performed and the similarity ratio with respect to the optimal is calculated.

A = D optimal for benefit(13)

A = optimal D for cost(14)

Step 4: The weighted normalization matrix (Z) is obtained by multiplying with the weights. Equation (15) is used in this process. In this context, w denotes the weights of the criteria.

Z = wA(15)

Step 5: The average similarity ratio is calculated by taking the geometric mean (G) of the weighted similarity ratios of each alternative. Equation (16) is used to achieve this goal.

G i = ∏ j = 1 n Z ij n(16)

Step 6: Average similarity rates are ranked. The highest value is considered like the most optimal.

MAIRCA method is one of the ranking methods in multi-criteria decision making. The method in question considers the gap between the theoretical evaluation matrix based on uniform probability distribution and the actual evaluation matrix [60]. In the analysis processes of this method, the decision matrix (D) is first created. After that, preference probabilities (P _{B i} ) are found for each alternative with the help of Equation (17) [61].

P B i = 1 m(17)

In the following stage, theoretical evaluation matrix (K_p) is obtained by multiplying the preference probabilities with weights. For this purpose, Equation (18) is taken into consideration [62].

K p = [ k p 11 ⋯ k p 1 n ⋮ ⋱ ⋮ k pm 1 ⋯ k pmn ] = [ P B 1 w 1 ⋯ P B 1 w n ⋮ ⋱ ⋮ P B m w 1 ⋯ P B m w n ](18)

Next, real evaluation matrix (K_r) is constructed. In this scope, Equation (19) is used where there are benefit criteria. However, for the cost criteria, Equation (20) is taken into consideration.

k r ′ j = k p ′ j ( d ' j - min ( d i ) max ( d i ) - min ( d i ) )(19)

k r ′ j = k p ′ j ( d ' j - max ( d i ) min ( d i ) - min ( d i ) )(20)

Moreover, by using Equation (21), the total void matrix (F) is calculated.

F = K p - K r = [ f 11 ⋯ f 1 n ⋮ ⋱ ⋮ f m 1 ⋯ f mn ] = [ k p 11 - k r 11 ⋯ k p 1 n - k r 1 n ⋮ ⋱ ⋮ k p m 1 - k r m 1 ⋯ k p mn - k r mn ](21)

Finally, the final score (U) for each alternative is calculated by Equation (22). The value with the lowest final score is determined as the best alternative.

U i = ∑ j = 1 n f ij(22)

4.1 Weighting the criteria

In this study, it is aimed to identify the key factors to minimize carbon emission problem. In the analysis process, firstly, the abstracts of 1711 studies in the “Sciencedirect” platform are obtained. After that, most frequent words are identified by considering N-gram methodology. While combining these results with the literature evaluations, five different indicators are selected to minimize carbon emission problem. The details of these items are given in Table 2.

Table 2 gives information about the five different factors that influence carbon emission. After that, an expert team is created with three different people. These people have at least 18 years of working experience. One of them is an academician that has lots of scientific publications about the carbon emission problem. On the other side, the other two people work as top managers in international energy companies. They have worked on many different projects regarding carbon emissions. Expert opinions are given in the appendix part (Table A1). Expert opinions are translated into fuzzy numbers with the values in Table 1. After that, the decision matrix (D) is obtained with the help of Equations (2). The decision matrix is presented in Table A2. In the next step, the normalized matrix (X) is obtained with the help of Equations (4). All criteria in the study are benefits and the normalized matrix is given in Table A3. Moreover, the index (C) value is calculated with the help of Equation (5)–(8) (Table A4). Finally, criterion weights (W) are calculated by Equation (9). Table 3 gives information about the weighting results.

It is concluded that renewable energy usage is the most critical item to overcome carbon emission problem because it has the greatest weight (.3013). On the other side, green buildings and technology are other key issues for this purpose. However, policies play a less important role in this context.

In this part of the proposed model, E7 countries are compared according to their carbon emission minimization performance. In this process, an analysis is carried out with the newly developed RATGOS model. The decision matrix (D) is formed by taking the arithmetic average of the expert opinions (Table A5). In the next step, the normalized decision matrix (A) is obtained by using Equation (11)-(14) (Table A6). Furthermore, by multiplying the normalized decision matrix with the weights obtained by Spherical fuzzy CRITIC, the weighted decision matrix (Z) is obtained with the help of Equation (15) (Table A7). Finally, G matrix is calculated by taking the geometric mean of each alternative as in Table 4.

Table 4 demonstrates that China is the most successful country to overcome carbon emissions problem. Similarly, it is also seen that Russia and Turkey also take meaningful actions in this framework. However, Indonesia and Mexico are on the last ranks.

In this stage, E7 countries are also ranked by using MAIRCA technique. First, the decision matrix is created from expert opinions. Then, the preference probability value is calculated by Equation (17) and the p value is found as 0.1429. Equation (18) obtained the theoretical evaluation matrix. For the weights in the equation, the weights obtained from the Spherical fuzzy CRITIC technique are considered. The theoretical evaluation matrix (Kp) is presented in Table A8. Then, the real evaluation matrix (Kr) is calculated (Table A9) with the help of Equation (19) since all criteria are benefit factors. Furthermore, with the help of Equation (21) the total void matrix (F) is calculated (Table A10). In the last step, U values are calculated with the help of Equation (22). These values are demonstrated in Table 5.

The lowest U score gives information about the best alternative. Hence, it is seen that China is the best country to cope with carbon emission problem. Figure 1 makes comparison about the ranking results of RATGOS and MAIRCA.

OPEN IN VIEWER

Fig. 1

Comparative ranking results.

Figure 1 denotes that the ranking results of RATGOS and MAIRCA are quite similar. Thus, it can be understood that the proposed model is reliable and coherent.

The RATGOS technique, which is considered in the ranking of E7 countries, is used for the first time in this study. Therefore, a sensitivity analysis has been performed to test the accuracy of this new technique. In this framework, the order in 10 different situations was calculated by taking the changes of 10%, 5%, and 1% of the weights obtained as a result of the analysis with Spherical fuzzy CRITIC. The analysis results obtained are given in Table 6.

Table 6 indicates that the analysis results are the same for all different cases. Therefore, it is understood that the findings of the proposed model are consistent and coherent.