Fuzzy clustering algorithm-based teaching resource recommendation method for automotive engine overhaul

Abstract

At present, the traditional teaching resource recommendation algorithm mainly constructs the evaluation matrix of teaching resources and determines the recommendation order according to the evaluation value, which lacks the grasp of user preferences and leads to poor recommendation effect. Therefore, a method of recommending teaching resources for automobile engine overhaul based on fuzzy clustering algorithm is proposed. This paper constructs the algorithm flow of automobile engine overhaul teaching resources recommendation, describes the K-means clustering algorithm and the improved K-means clustering algorithm, analyzes the user characteristics by using the clustering fuzzy algorithm, calculates the similarity between user interests and resources, and completes the recommendation of automobile engine overhaul teaching resources. Experiments verify the recommendation accuracy of this method, and the results show that this method has low MAE value and high recommendation accuracy when it is used to recommend teaching resources.

Keywords

fuzzy clustering algorithm automotive engines teaching resources recommendation algorithm

Introduction

With the rapid development of automobile industry, automobile engine maintenance technology, as an important part of automobile aftermarket service, has attracted more and more attention.¹ In order to improve the training quality of automobile engine maintenance talents and optimize the allocation of teaching resources, based on fuzzy clustering algorithm, the recommendation method of automobile engine maintenance teaching resources is discussed.² Through the in-depth excavation and intelligent analysis of existing teaching resources, it aims to realize the optimal allocation of teaching resources and improve the teaching effect and talent training quality in the field of automobile engine maintenance.

In recent years, fuzzy clustering algorithm has been widely used in many fields, which has obvious advantages in dealing with fuzzy and uncertain data. In the teaching process of automobile engine maintenance, the selection and allocation of teaching resources are influenced by many factors, such as students’ foundation, teachers’ experience, and equipment conditions. There are certain fuzziness and uncertainty among these factors.³ Therefore, this study attempts to introduce fuzzy clustering algorithm into the field of automobile engine maintenance teaching resource recommendation, in order to provide useful reference for practical teaching work.

With the continuous advancement of educational informatization and the wide application of big data technology,⁴ the teaching resource recommendation method based on fuzzy clustering algorithm is expected to play an increasingly important role in the future education field. This study will take this opportunity to further promote the application and development of fuzzy clustering algorithm in the field of education and make contributions to improving the quality and efficiency of education. This study not only helps to enrich the application fields of fuzzy clustering algorithm and expand its application prospects in the field of education but also provides scientific basis and practical guidance for the recommendation of teaching resources for automobile engine maintenance. Through this study, it can provide useful reference for automobile engine maintenance teaching workers, promote the optimal allocation of automobile engine maintenance teaching resources, and improve the quality of personnel training. At the same time, this study will also provide reference and enlightenment for the application of fuzzy clustering algorithm in other educational fields.

In a word, the recommendation method of teaching resources for automobile engine maintenance based on fuzzy clustering algorithm in this study has important theoretical significance and practical value. Through in-depth mining and analysis of teaching resource data, intelligent recommendation and optimal allocation of teaching resources are realized, which provides beneficial support for automobile engine maintenance teaching and promotes the improvement of talent training quality. At the same time, this study will also provide new ideas and methods for the application and development of fuzzy clustering algorithm in the field of education.

User feature analysis based on fuzzy clustering algorithm

Traditional collaborative filtering algorithms generally recommend users based on the rating matrix only,⁵ and the specific model structure is shown in Figure 1.

Figure 1.

Traditional recommendation algorithm flow.

In the teaching engine overhaul teaching resources system, the user rating data is small compared to the scale of engine overhaul teaching resources, so the overall rating data is sparse. If two users with similar interests in the system do not have a common rating engine overhaul teaching resources, it is not possible to make recommendations for one user based on the rating data of another user, so the data sparsity problem is an important factor affecting the recommendation quality of the algorithm.⁶ For this problem, this paper considers the attribute characteristics of engine overhaul teaching resources. In general, when users generally rate engine overhaul teaching resources with a certain type of attribute higher, then it can be assumed that this user has a preference for this type of attribute.⁷ The similarity of users can be measured through a combination of rating similarity and similarity of engine overhaul teaching resource attribute preference, which can explore the potential points of interest of users, increase the density of data, and alleviate the problem of data sparsity. The flow chart of the recommendation algorithm proposed in this paper is shown in Figure 2.

Figure 2.

Algorithm flow for recommending teaching resources of engine overhaul for automotive engine overhaul.

When new users first enter the recommendation system, there is no evaluation data on engine overhaul teaching resources, so it is difficult to use the rating to recommend users, which creates the problem of user cold start. However, new users have certain characteristic information, such as age, gender, and profession. This characteristic information represents the user’s own attributes, which can more truly reflect the relationship between users and users and between users and engine overhaul teaching resources.⁸ Therefore, in this paper, user features are used to measure the similarity degree of new users and obtain the nearest neighbors of users. And information entropy is introduced to measure the amount of information contained in user ratings, and the information entropy of users is calculated in the cluster where the nearest neighbor users are located, and the ratings of new users are predicted based on the ratings of users with higher information entropy, which improves the recommendation quality of new users and improves the user cold start problem.

When recommending users, we first classify users with the same interest and preference into the same cluster according to the clustering algorithm, and then select a number of clusters close to the target user and perform nearest neighbor search in them, which realizes “cross-category” recommendation and reduces the time and space overhead of search, and the clustering process can be performed offline.

Fuzzy cluster analysis is an important data analysis technique that has a wide range of applications in pattern recognition, image processing, and document classification. Cluster analysis refers to the analysis process of grouping a collection of physical or abstract objects into multiple classes consisting of similar objects. The goal of clustering is to divide the data into multiple distinct clusters, and the division is based on the magnitude of similarity between objects, that is, as much similarity as possible between objects in each cluster and as little similarity as possible between objects in different clusters.⁹

In collaborative filtering recommendation techniques, the sparsity of the scoring matrix is the key to the problem, so it is necessary to deal with the representation of the user-resource scoring matrix accordingly. The potential content similarity relationship between resources cannot be ignored, and resources that are similar in content imply that users will have the same interest orientation to such resources, so it is not necessary to look for similar interests of users on each resource but rather on a set of resources. In this paper, we first apply the fuzzy clustering technique to cluster the resources in terms of their attribute features.

The similarity measure between objects is the key to clustering, and its accuracy directly affects the effect of clustering. The smaller the distance, the greater the similarity between the objects; and the larger the similarity coefficient, the greater the similarity between the objects.¹⁰ Let an object have n attributes, then it can be represented by an n-dimensional vector, such as the objects $X_{i}$ and $X_{j}$ can be represented as $X_{i} = (x_{i 1}, x_{i 2}, . . ., x_{i m})$ and $X_{j} = (x_{j 1}, x_{j 2}, . . ., x_{j m})$ , and then their Euclidean distance calculation formula is shown below.

d_{i j} = {(\sum_{k = 1}^{n} {| x_{i k} - x_{j k} |}^{2})}^{1 / 2}

(1)

In this paper, use K-means clustering algorithm to cluster and analyze the user characteristics of teaching engine overhaul teaching resources. K-means algorithm is a division-based clustering algorithm, which is very widely used at present and has greater advantages in big data processing. The algorithm starts by subjectively specifying K initial clustering centers, first roughly dividing users into these clusters for the first time, then iteratively modifying the classification according to some optimization principles, and finally obtaining the classification with higher similarity of users within the cluster.¹¹

The brief process of K-mean: algorithm can be outlined as follows: firstly, K-objects are randomly selected from the object dataset as the initial cluster centers of prime points, and the remaining objects are divided into K-clusters according to the similarity with the cluster centers, and the mean centroids of the objects in each cluster are used to replace the original cluster centers and iterate continuously until convergence.¹² The specific algorithm description is shown in Table 1.

Table 1.

Description of K-means clustering algorithm.

Algorithm steps	Algorithm description
Input	Set of data objects, number of clusters K
Output	K-clusters
Algorithm steps	Randomly select K initial cluster centroids from the set of data objects
	Calculate the distances of the remaining objects to the initial K centroids and divide them into clusters with the smallest distance from the cluster center
	Calculate the mean value of all objects in the cluster to get the new cluster center
	Repeat steps 2 and 3 until the cluster centers are stable and do not change

A standard measurement function is generally used to determine convergence, usually by calculating the mean squared deviation, as shown below.

E = p - m_{i}

(2)

In formula (2), $E$ represents the sum of mean squared deviations of all objects, $p$ represents a node in the object space, and $m_{i}$ represents the cluster centroid. The specific algorithm flow is shown in Figure 3.¹³

Figure 3.

Flow chart of K-means clustering algorithm.

The advantages of the K-means algorithm are that the algorithm is simple to implement, efficient, and scalable, and the clustering effect is better in the case of relatively dense objects. However, the algorithm also has some limitations, the number of clusters K needs to be specified artificially, the selection of K-value is difficult to determine, and most of the time it is not sure how many clusters are most appropriate to divide the data objects into.¹⁴ In addition, the initial clustering centers are randomly selected to have a large impact on the clustering results, and the algorithm iteratively solves for the optimal value based on the initial clustering centers, but the algorithm is not globally optimal but locally optimal. The clustering results are not stable, and different initial values may lead to different clustering results, so the quality of clustering depends on the initial values, and the determination of the initial clustering centers is very critical.¹⁵

To address the problems in the K-means algorithm, the initial clustering centers are determined based on the maximum distance, which makes the initial clustering center selection follow a certain pattern, reduces the influence of random selection on the clustering results, and clusters the users according to the preference model of engine overhaul teaching resource attributes. The specific description of the improved clustering algorithm is shown in Table 2.¹⁶

Table 2.

Description of the improved K-means algorithm.

Algorithm steps	Algorithm description
Input	Initial number of clusters K, rating information table, and resource attribute table
Output	K-clusters
Algorithm steps	Retrieve the set of all users from the rating information table
	Retrieve all resource collections from the Resource Properties table
	Transform resource attribute table and rating information into resource attribute matrix and rating matrix
	Calculate user preferences for resource attributes
	Calculate user distance
	Calculate the clusters of remaining users to the initial center and classify the users into the clusters with the smallest distance

According to Table 2, firstly, the input of the algorithm includes initial cluster number k, scoring information table, and resource attribute table, which are used for the operation and calculation of the algorithm. Secondly, the output of the algorithm is k-clusters, that is, users are divided into k different groups by clustering algorithm. In terms of the steps of the algorithm, it includes obtaining the set of all users from the scoring information table, obtaining all resource sets from the resource attribute table, transforming resource attributes and scoring information into resource attribute matrix and scoring matrix, calculating users’ preferences for resource attributes, calculating the distance between users, calculating the cluster of remaining users to the initial center, and classifying users into the cluster with the smallest distance. The improved K-means algorithm first selects K initial clustering centers with the farthest distance based on the distance between users and then iteratively clusters the users according to the process of the traditional K-means algorithm. The distance measure among users is adopted by the Euclidean distance formula based on the preference of engine overhaul teaching resources attributes.¹⁶

User similarity metric

User similarity measurement is based on the analysis of user characteristics. By comparing the similarity between users in certain characteristics or dimensions, we can measure their similarity in common interests and needs. This similarity measure can be based on various measurement methods, such as Euclidean distance, cosine similarity, and so on. By calculating the similarity measure, other users who are most similar to the target users can be found, and their common interests and needs can be obtained from their behaviors, which is helpful to provide personalized recommendations, establish user social networks, and promote communication and interaction between users.

There are various ways to calculate the similarity between users’ interests and resources. The product of item similarity and rating is used as the similarity between users' interests and resources; content-based recommendation uses the cosine similarity of text as the similarity between users’ interests and resources. Different recommendation methods have different methods for calculating the similarity between user interests and resources.¹⁷

In this paper, due to the limitation of user feature data in the dataset, the keywords in the text are used as user interest features, and the cosine similarity between the semantic features of the historical resources that learners have learned and the semantic features of the engine overhaul teaching resources is calculated as the similarity between user interest and resources.¹⁸ Text is a high-dimensional semantic space that cannot be directly received by the computer and needs to be represented as a word vector. In this paper, word2vec is used for word vector representation of text data. word2vec uses a neural network model to represent the text content as vectors on a K-dimensional vector space, and the semantic similarity of the text is represented by calculating the similarity on the vector space. The resource text data is divided into words, deactivated words and other operations are removed, and keywords are extracted using the TextRank algorithm. In this paper, the words are represented as 54-dimensional vectors using word2vec.

After completing the word vector representation of the resource, the word vector is used to calculate the similarity of the text as the similarity between the user’s interest and the resource.¹⁹ The similarity is defined as follows:

I_{n t e r e s t} (u, z) = \frac{1}{| N (u) |} \times \bar{w_{i}} \times \bar{w_{j}}

(3)

In formula (3), $u$ represents the set of users, $z$ represents the target user, $N (u)$ denotes the set of historical engine overhaul teaching resources of this user, $| N (u) |$ denotes the number of the set of historical engine overhaul teaching resources of the user, and $\bar{w_{i}}$ and $\bar{w_{j}}$ denote the word vector representations of resource $i$ and resource $j$ , respectively. In the calculation of similarity, the higher the textual similarity between the resource and the user’s historical resources, the more similar the resource is to the user’s preferences, the more recommendable the resource is.

In this paper, due to the limitation of the dataset, text similarity is used to represent the similarity between user interests and resources. In case of more data, this section can also be used to model user interests using more features, using other methods such as LFM and LDA topic model to calculate user preferences.²⁰

In the same application field, the preferences of the same user for similar engine overhaul teaching resources in terms of attribute characteristics should be similar, so the engine overhaul teaching resources that are similar in terms of attribute characteristics can form a large category, and the user’s preference for this large category represents the user’s preference for all the engine overhaul teaching resources contained in the large category. To construct the user’s evaluation preference for the category, one is to reduce the dimensionality of the user-engine overhaul teaching resource preference matrix, because the number of clusters is much smaller than the number of engine overhaul teaching resources, and the calculation of the low-dimensional matrix must take less time than that of the high-dimensional matrix, thus improving the scalability of the recommendation method; secondly, user preferences for similar engine overhaul teaching resources are largely consistent; third, the evaluation preference of user one engine overhaul teaching resources is sparse while the evaluation preference of user one fuzzy cluster is dense, which can solve the problem of inaccurate similar group metrics caused by the sparsity of evaluation data.

The fuzzy clusters will form corresponding fuzzy clusters after fuzzy clustering, and the affiliation degree of each engine overhaul teaching resource relative to a fuzzy cluster will be calculated using formula (4).²¹

μ_{i} (x_{i}) = \frac{‖ x_{i} - m_{j} ‖}{\sum_{k = 1}^{c} m_{k}}

(4)

In formula (4), $x_{i}$ represents the attribute feature vector of each engine overhaul teaching resource, $m_{k}$ represents the attribute feature vector of each fuzzy cluster center, $‖ x_{i} - m_{j} ‖$ represents the proximity of engine overhaul teaching resource $i$ to fuzzy cluster $j$ , and $c$ represents the engine overhaul teaching resource of fuzzy cluster.

The dense user-a-fuzzy cluster preference matrix shown in Table 3 is constructed from the preference values expressed by users in the user-engine overhaul teaching resource sparse evaluation preference matrix and the affiliation degree of each engine overhaul teaching resource belonging to each fuzzy cluster, and the values in the matrix represent the evaluation preference values of user u for cluster j.

Table 3.

Preference matrix of user-fuzzy clusters.

User-cluster	C1	C2	…	Cc
U1	4	3	…	1
U2	5	2	…	1
…	…	…	…	…
Un	4	3	…	1

Where the formula for calculating the preference value is shown below.

C_{u, j} = \frac{I_{u, i} \times P_{u}}{μ_{j}}

(5)

In formula (5), $I_{u, i}$ represents the evaluation preference value of user $u$ for fuzzy cluster $i$ , $μ_{j}$ represents the preference value of user $u$ for engine overhaul teaching resource $j$ , and $P_{u}$ represents the set of engine overhaul teaching resources that have been evaluated by user $u$ .

Recommendation result generation

The ultimate goal of the collaborative filtering algorithm is to give the user’s predicted rating and generate recommendation results based on the predicted rating. The above two sections construct a user-engine overhaul teaching resource rating matrix and a user-category rating matrix, respectively. Both matrices reflect the actual situation of a user in terms of subjective rating and preference degree, and both must be considered in order to predict results that fit the actual situation. The main idea is to find out the clusters to which the target user belongs according to the results of user clustering, calculate only the similarity between the target user and other users in the clusters, and arrange them in descending order of similarity, with the top-ranked users forming the neighbor set of the target user.²²

Taking the target user u as an example, for u and one of his neighbors v, the user-engine overhaul teaching resource rating similarity $g_{s i m} (u, v)$ and the category rating similarity $c_{s i m} (u, v)$ are calculated using the Pearson correlation coefficient method, respectively. The final similarity formula can be obtained by weighting and summing the two, as shown in formula (6). $λ$ is the weighting factor, which is used to adjust the weight of user rating similarity in the overall user similarity, and takes the value between [0, 1].²³

s_{u m} (u, v) = λ \times g_{s i m} (u, v) + (1 - λ) c_{s i m} (u, v)

(6)

The user with the highest similarity is selected as the set of nearest neighbors of u, denoted as $N_{u}$ , and then the predicted rating of user u for engine overhaul teaching resource j is calculated by formula (7). $r_{v j}$ denotes the predicted rating of user $v$ on engine overhaul teaching resource $j$ , ${\bar{r}}_{v}$ denotes the average rating of user $u$ on engine overhaul teaching resource, and $r_{u}$ denotes the average rating of user $v$ on engine overhaul teaching resource. If a neighbor exists that has no rating for engine overhaul teaching resource $j$ , then that neighbor user does not participate in the prediction of user $u$ 's rating for $j$ .²⁴

r_{u j} = r_{u} + \frac{\sum_{v \in N u} s_{i m} (u, v) \times (r_{v j} - {\bar{r}}_{v})}{\sum_{v \in N u} s_{i m} (u, v)}

(7)

The above formula (7) predicts the engine overhaul teaching resource rating based on the user rating level ${\bar{r}}_{v}$ , then predicts the difference of the neighbor’s rating on the engine overhaul teaching resource relative to their own rating level by fusing their similarity to the target user, and finally predicts the target user’s scoring. The scores are sorted from largest to smallest, and the front-end engine overhaul teaching resources are recommended to the users. So far, the construction of the fuzzy clustering algorithm based automotive engine overhaul teaching resources recommendation method is completed.

Testing and analysis

Test preparation

In order to prove that the fuzzy clustering algorithm-based teaching resource recommendation method for automotive engine overhaul proposed in this paper is better than the conventional teaching resource recommendation method in terms of overhaul effect, after the design of the theoretical part is completed, an experimental session is constructed to test the evaluation effect of the method in this paper. In order to improve the reliability of the experimental results, in addition to the method of this paper, two conventional methods were also selected as the experimental control group, which are the teaching resource recommendation method based on user rating and the teaching resource recommendation method based on knowledge graph.

In order to verify the effectiveness and recommendation quality of the recommendation method proposed in this paper, the experimental dataset uses a city basic education teaching resource library. The resource content covers nearly 10TB capacity of all subjects in basic education, with more than 9000 registered users and more than 1000 daily visits. Through collation, 31,000 interest degree evaluation data of 1600 teaching resources by 300 users were collected, and the sparsity level of the dataset is 1–31,000/(300*1600) = 0.9354. The evaluation value is an integer from 0 to 5, and the higher the value, the higher the user’s interest in the teaching resource is favorably evaluated, and the configuration of the experimental environment is shown in Table 4.

Table 4.

Experimental environment configuration.

Experimental configuration	Specific parameters
Deploy applications	Spark
CPU	16G
Memory	256G
Operating system	64-Ubuntu Sever 14.04
System Software	JDK1.7

The above three recommendation algorithms are used to recommend users and compare the effectiveness of the recommendation methods.

Analysis of test results

The comparison criterion chosen for this experiment is the recommendation accuracy of the method, and the specific measure is the mean absolute deviation (MAE), which represents the absolute value of the deviation between the actual value and the predicted value of the user’s interest in teaching resources, and the lower the value represents the higher the recommendation accuracy of the recommendation algorithm, and the specific calculation formula is shown below.

M A E = \sum_{i = 1}^{n} \frac{| P_{i} - Q_{i} |}{n}

(8)

In formula (8), $P_{i}$ represents the actual value of users’ interest in teaching resources and $Q_{i}$ represents the predicted value of the recommendation algorithm, and the specific experimental results are shown in Figure 4.

Figure 4.

Comparison of MAE values.

From the experimental results, the proposed fuzzy clustering-based teaching resource recommendation method has a smaller MAE value, and thus the recommendation quality of this method is better than that of two traditional recommendation algorithms. In addition, through testing and analysis of the experimental dataset, it is found that the fuzzy clustering-based collaborative filtering recommendation method takes into account the similarity of the recommended resources in terms of attribute characteristics, which improves the “cold start” problem faced by the collaborative filtering recommendation algorithm and therefore improves the recommendation quality of the recommendation algorithm.

Conclusion

This paper puts forward a recommendation method to improve the teaching resources of automobile engine maintenance by using fuzzy clustering algorithm. Through in-depth study, fuzzy clustering algorithm is used to recommend teaching resources to provide more effective learning experience. Through this study, the following conclusions are drawn: this method improves the “cold start” problem and brings the similarity of resource attributes into the similarity measurement of collaborative filtering recommendation method, which makes the correlation between resources more comprehensive and improves the recommendation accuracy of recommendation system.

In view of the above conclusions and research results, the following suggestions are put forward for the future. First of all, further improve and optimize the application of fuzzy clustering algorithm in the recommendation of teaching resources for automobile engine maintenance. By introducing more features and indicators, the accuracy and efficiency of the algorithm are optimized to provide more accurate and personalized resource recommendation. Secondly, encourage the development of personalized learning systems based on artificial intelligence and machine learning, and provide students with more personalized learning paths and resource recommendations on the basis of a comprehensive understanding of students’ needs. Finally, strengthen cooperation with the practice of automobile engine maintenance, and collect more actual data and cases to verify and improve the proposed fuzzy clustering algorithm. In the process of recommending teaching resources, interactive and feedback mechanisms are added to continuously optimize the recommendation results and make adjustments and improvements according to students’ feedback.

Statements and declarations

Footnotes

Conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Academic funding project for Science and Technology Research Program of Chongqing Education Commission of China under Grant No. KJZD-K202404001.

References

Corno

D'Avico

Marelli

, et al. Predictive cylinder deactivation control for large displacement automotive engines. IEEE Trans Veh Technol 2019; 68(10): 9554–9563.

Laohakiat

Sa-Ing

. An incremental density-based clustering framework using fuzzy local clustering. Inf Sci 2021; 547: 404–426.

Chen

Wang

, et al. Dynamic reactive power optimization of distribution network with distributed generation based on fuzzy time clustering. IET Gener Transm Distrib 2022; 16(7): 1349–1363.

Tang

, et al. Fuzzy c-means clustering based segmentation and the filtering method for discontinuous ESPI fringe patterns. Appl Opt 2019; 58(6): 1442–1450.

Trinh

Huynh

Bidaki

, et al. Optimized fuzzy clustering using moth-flame optimization algorithm in wireless sensor networks. Artif Intell Rev 2021; 55(3): 1915–1945.

Kulkarni

Jena

Sankar

. MapReduce framework based big data clustering using fractional integrated sparse fuzzy C means algorithm. IET Image Process 2020; 14(12): 2719–2727.

Liu

Lin

Wen

, et al. Fuzzy equivalence relation clustering-based algorithm for coherency identification among generators. J Energy Eng 2019; 145(1): 04018070.

Urroz

Diéguez

Arzamendi

, et al. Gaseous fueling of an adapted commercial automotive spark-ignition engine: Simplified thermodynamic modeling and experimental study running on hydrogen, methane, carbon monoxide and their mixtures. Fuel 2023; 337: 127178.

Mojarad

Nejatian

Parvin

, et al. A fuzzy clustering ensemble based on cluster clustering and iterative Fusion of base clusters. Appl Intell 2019; 49(7): 2567–2581.

10.

Zhu

. Evaluation of innovation capability based on fuzzy C-means clustering of user evaluation criteria. Basic Clin Pharmacol 2019; 125(9): 178–179.

11.

Lohani

QMD

Solanki

Muhuri

. Novel adaptive clustering algorithms based on a probabilistic similarity measure over atanassov intuitionistic fuzzy set. IEEE Trans Fuzzy Syst 2018; 26(6): 3715–3729.

12.

Zhang

Yue

, et al. Interval type-2 fuzzy local enhancement based rough k-means clustering considering imbalanced clusters. IEEE Trans Fuzzy Syst 2019; 28(9): 1925–1939.

13.

Morán

Fuentes

Liu

, et al. FracVAL: An improved tunable algorithm of cluster-cluster aggregation for generation of fractal structures formed by polydisperse primary particles. Comput Phys Commun 2019; 239: 225–237.

14.

Calder

Roy

Wang

. Performance and emissions of a diesel engine fueled by biodiesel-diesel blends with recycled expanded polystyrene and fuel stabilizing additive. Energy 2018; 149: 204–212.

15.

Dang

Wang

. Optimization control of hydrogen engine ignition system based on ACO-BP. Int J Hydrogen Energy 2021; 46(78): 38903–38912.

16.

Osipov

Fadin

Nikanorov

. Wear and coefficient of friction of a supermodified hypereutectic aluminum-silicon alloy. Tech Phys 2020; 65(12): 1981–1986.

17.

Chen

Wang

, et al. Integrating User-Group relationships under interest similarity constraints for social recommendation. Knowl Base Syst 2022; 249: 108921.

18.

Xia

Zhou

Lin

. Online recursive power management strategy based on the reinforcement learning algorithm with cosine similarity and a forgetting factor. IEEE Trans Ind Electron 2021; 68(6): 5013–5023.

19.

Boudewyn

Kiat

, et al. Neural correlates of word representation vectors in natural language processing models: Evidence from representational similarity analysis of event-related brain potentials. Psychophysiology 2022; 59(3): 13976.

20.

Yan

. A Siamese neural network for learning the similarity metrics of linear features. Int J Geogr Inf Sci 2023; 37(3): 684–711.

21.

Fang

Yang

, et al. New feature analysis-based elastic net algorithm with clustering objective function. Knowl Base Syst 2022; 258: 110004.

22.

Ryu

Wang

Pak

, et al. A clustering based traffic flow prediction method with dynamic spatiotemporal correlation analysis. Transportation 2022; 49(3): 951–988.

23.

Wen

Xuan

, et al. Image segmentation algorithm based on neutrosophic fuzzy clustering with non-local information. IET Image Process 2020; 14(3): 576–584.

24.

Bui

Snasel

, et al. SFCM: A fuzzy clustering algorithm of extracting the shape information of data. IEEE Trans Fuzzy Syst 2021; 29(1): 75–89.