Analysis and Evaluation of the Relationship between Teaching Pressure and Self-Efficacy of College Teachers Based on Artificial Neural Network

Wu, Songjiao

doi:https://doi.org/10.1155/2022/3755257

Scientific Programming

On this page

Abstract Introduction Conclusions Data Availability Conflicts of Interest References Copyright Related Articles

Special Issue

Scientific Programming for Industry 5.0: Theory, Applications, and Technological Development

View this Special Issue

Research Article | Open Access

Volume 2022 | Article ID 3755257 | https://doi.org/10.1155/2022/3755257

Analysis and Evaluation of the Relationship between Teaching Pressure and Self-Efficacy of College Teachers Based on Artificial Neural Network

Songjiao Wu¹

Academic Editor: Ahmed Farouk

Received20 Jan 2022

Revised01 Apr 2022

Accepted04 Apr 2022

Published28 Apr 2022

Abstract

With the reform of the education system, the society today raises higher requirements for college teachers, which cause immense psychological stress among them. To enhance the quality of teachers, it is important to analyze the relationship between teaching pressure and self-efficacy. Therefore, this paper tries to analyze and evaluate the relationship between teaching pressure and self-efficacy of college teachers based on artificial neural network. Firstly, a grey correlation analysis (GRA) model was established for the teaching pressure and self-efficacy of college teachers, and the analysis procedure was detailed. Then, the possible multicollinearity of the GRA model was tested. In addition, a linear regression model was established based on Lasso variable selection model and ridge regression variable selection model, aiming to eliminate the multicollinearity between various teaching pressure factors in the GRA model. Finally, a multilabel learning algorithm was proposed based on neural network and label correlation. In this way, the correlations between the various teaching pressure factors and teachers’ self-efficacy were mined automatically. The proposed model proved valid through experiments.

1. Introduction

With the reform of the education system, the society today raises higher requirements for college teachers: a college teacher needs to play multiple roles at the same time, namely, knowledge imparter, students’ psychological tutor, class leader, and pioneer of advanced teaching method [1–5]. Many college teachers find it difficult to strike the balance between being an ordinary person and acting as a role model, and thus face an immense psychological stress [6–10]. Survey results show that college teachers of the same level, although faced with the same pressure in the same period, could be optimistic or pessimistic. When a teacher perceives a low teaching pressure, he/she will work actively, have a high self-efficacy, and improve his/her qualities rapidly. Therefore, it is important to analyze the relationship between teaching pressure and self-efficacy [11–13].

Chung and Chen [14] compared the self-efficacy, job satisfaction, and pressure of teachers in application-oriented colleges in Fujian and Taiwan and discussed the role of self-efficacy in this context. Considering the importance of science and educational psychology in education system, Kuo et al. [15] treated learning motivation as the predictor variable and self-efficacy as the evidence variable and tried to discover the important correlation of the learning motivation and self-efficacy of teachers and students with the principle of educational psychology. Hamedi [16] pointed out that education informatization is an inevitable trend of higher education reform; identified teachers as the key to advancing and applying information education; attributed the psychological stress of college teachers to subjective and objective sources under education informatization; and suggested that teachers should adjust their cognition, emotions, will, and behaviors to meet the needs of education. Wang and Wang [17] revised the technology acceptance model to focus on three individual differences (self-efficacy, personal innovation ability, and sensitivity to environmental pressure).

The above is a brief review of the research into the teaching pressure and self-efficacy of college teachers. It can be seen that the studies at home and abroad have achieved a lot of results, but some defects are yet to be solved. In terms of research contents, the relevant research is immature, failing to fully consider the various indices of teaching pressure. In terms of research methods, most studies rely on questionnaire survey and qualitative research. The scientific level and rigor must be improved by introducing artificial intelligence (AI) strategies. Therefore, this paper tries to analyze and evaluate the relationship between teaching pressure and self-efficacy of college teachers based on artificial neural network [18–20]. Section 2 establishes a grey correlation analysis (GRA) model for the teaching pressure and self-efficacy of college teachers and details the analysis procedure. Besides, the possible multicollinearity of the GRA model was tested with Pearson correlation coefficient and variance inflation factor (VIF). Then, a linear regression model was established based on Lasso variable selection model and ridge regression variable selection model, aiming to eliminate the multicollinearity between various teaching pressure factors in the GRA model. Section 3 proposes a multilabel learning algorithm based on neural network and label correlation and relies on the algorithm to automatically mine the correlations between the various teaching pressure factors and teachers’ self-efficacy. The proposed model proved valid through experiments.

2. Multifactor Correlation Analysis Model

2.1. Model Construction

In recent years, domestic and foreign scholars have achieved fruitful results on the correlation analysis of different variables. The main tools used for correlation analysis are GRA, least squares regression, etc.

The foreign research of self-efficacy began with the American psychologist Albert Bandura, who created the concept of self-efficacy in 1977. Seven years later, he defined self-efficacy from the angle of social cognitive theory as an individual's belief in their capacity to execute behaviors necessary to produce specific performance attainments. The self-efficacy is mainly influenced by six factors: performance experience, vicarious experience, imaginal experience, social persuasion, physical arousal, and psychological state. To derive the correlation between college teachers’ teaching pressure and self-efficacy, it is important to exclude the unimportant factors and sort the influencing factors by the degree of correlation. For this purpose, the study constructs a GRA model. The analysis procedure of the proposed model is as follows.

Step 1. To reflect the systematic correlations and fully consider the possible multicollinearity, it is necessary to set up the reference series and comparative series for correlation analysis. This paper takes the quantified self-efficacy of teachers as the reference series Q and the quantified values of teaching pressure factors as the comparative series W_i. Let l be the calculation moment and i be the number of rows of influencing factors. Then, the reference series and comparative series can be, respectively, expressed as

Step 2. Preprocess the collected sample data; i.e., take the average of the series data of each teaching pressure factor:Let τ ∈ [0, 1] be the resolution coefficient that determines the difference of correlation coefficients. Then, the correlation coefficient between self-efficacy of college teachers and each teaching pressure factor can be calculated by

Step 3. Compute the correlation degree between self-efficacy of college teachers and each teaching pressure factor. Here, correlation coefficient is introduced to measure the degree of correlation between the two concepts. The correlation coefficient has various values, which correspond to the influencing factors. To reduce the effects of dispersion of sample data on overall correlation comparison, this paper takes the average of the correlation coefficients corresponding to different influencing factors at different moments. The correlation degree can be calculated byWhen the correlation degree e falls in [0, 0.25], teachers’ self-efficacy has a low correlation with teaching pressure factors. When e falls in [0.25, 0.5], teachers’ self-efficacy has a medium correlation with teaching pressure factors. When e falls in [0.5, 0.75], teachers’ self-efficacy has a relatively strong correlation with teaching pressure factors. When e falls in [0.75, 1], teachers’ self-efficacy has a highly strong correlation with teaching pressure factors.

Step 4. After computing the correlation degree of each teaching pressure factor, rank the various factors by the correlation degree with teachers’ self-efficacy.
To test the possible multicollinearity of the GRA model, the teaching pressure factors selected by the model were tested based on Pearson correlation coefficient.
The author firstly calculated the covariance XF(, q) = HO(W,Q)-HO(W)HO(Q) between each influencing factor and teachers’ self-efficacy and then computed the standard deviation of the influencing factors and that ε_q of teachers’ self-efficacy. Let HO be the expectation. Then, the Pearson correlation coefficient between each teaching pressure factor and teachers’ self-efficacy can be calculated byIf = 0, the teaching pressure factor is not linearly correlated with teachers’ self-efficacy; if > 0, the two have a positively correlation; if < 0, the two have a negative correlation; if > 0.8, the two have a very strong linear correlation.
Next, the VIFs were calculated for the teaching pressure factors. The first step is to compute the coefficient of multiple determination of the current teaching pressure factor relative to the other influencing factors. The VIF can be calculated byIf the VIF > 100, the GRA model faces a severe multicollinearity between the various teaching pressure factors. If the VIF is greater than 10 and smaller than 100, the model factors have a relatively strong multicollinearity. If the VIF is greater than 0 and smaller than 10, the model factors have a negligible multicollinearity. Figure 1 shows the execution flow of the multifactor correlation analysis model, which consists of model construction and multicollinearity test.

2.2. Variable Selection

Our regression model was constructed based on Lasso variable selection model, aiming to eliminate the multicollinearity between the various teaching pressure factors of the GRA model.

Let Q be teachers’ self-efficacy, W be the matrix of teaching pressure factors, α be the parameter to be estimated, and σ be the error term. For a general linear regression model, there is Q = Wα + σ. After centralizing Q and normalizing W, the least squares estimation can be expressed as

The parameter to be estimated satisfies

Under the constraint ∑ⁿ|α| ≤ p, the Lasso regression can be derived from formula (7):

Let μ be a positive penalty parameter controlling the number of influencing factors. The value of this parameter can be computed through cross validation. Based on Lagrangian duality, the above formula can be converted into

The greater than penalty parameter, the closer the regression coefficient of teaching pressure parameters to zero.

This paper also constructs a ridge regression variable selection model, aiming to eliminate the multicollinearity between influencing factors and form a contrast against Lasso variable selection model. For a general linear regression model, there is

By least squares method, the regression coefficient α can be estimated as

The regression coefficient of ridge regression can be solved by

Let l ∈ [0, 1] be the ridge regression coefficient. The greater the value of l, the smaller the stability of the regression parameter to be affected by the multicollinearity between influencing factors, and the larger the variance of the predicted correlation. To mitigate the influence of variable dimensionality on predicted correlation, the teaching pressure factors of the model must be normalized before ridge regression. Let o_ij and e_ij be the values of original and normalized factors, respectively, and λ_j and ε_j be the arithmetic mean and standard deviation of variable j, respectively. Then, we have

Figure 2 shows the flow of constructing linear variable regression models based on Lasso variable selection model and ridge regression variable selection model.

3. Multilabel Correlation Analysis

To automatically mine the correlations between teaching pressure factors and teachers’ self-efficacy, this paper proposes a multilabel learning algorithm based on neural network and label correlation. The algorithm design mainly includes making reasonable use of the underlying correlation between teaching pressure factors, pruning additional features, exploring the correlation degree of the labels corresponding to the influencing factors, and reconstructing the feature set of factor attributes. Figure 3 shows the flow of the multilabel learning algorithm.

During the training of the learning algorithm, a binary classifier is firstly trained for the label corresponding to each influencing factor, to obtain the predicted label (1 ≤ i ≤ x) of each sample. Then, is combined with each sample feature to form the augmented feature sets of training samples and test samples ( and ). Based on , the class A_i(1 ≤ i ≤ y) is obtained through label training. Finally, label (1 ≤ i ≤ x) is predicted for based on A_i. The additional feature information of each label can be expressed as

The additional feature information of each influencing factor is pruned to lower the probability of noise of the additional information, reduce the dimensionality of label information, and simplify the entire algorithm. Figure 4 shows the flow of pruning. Out of the 249 teachers selected for this research, 205 provided effective responses.

Firstly, the original training samples are divided into a training set C^TR and a verification set C^VE in a ratio of 8 : 2. Next, a binary classifier is trained on the training set C^TR, and the binary classifier corresponding to the label k_j of the j-th factor is denoted as A_j. Then, the label set of C^VE is verified. F1 is introduced to measure the hardness of the label:

This paper adopts the following pruning formula to judge whether the label is prone to prediction failure:

Let Ψ be the preset threshold of F₁; F(A_j,, C^VE) be the F₁ of k_j; and Ψ_j be the threshold. If Ψ_i > Ψ_j, the label k_j of the j-th factor has a high confidence; if Ψ_i < Ψ_j, k_j is error-prone and should not be adopted as additional information.

Considering the inconsistency between the labels corresponding to different factors, this paper directly uses the predicted labels of these factors to train the label learning algorithm. The additional feature set of the influencing factors is constructed on RBF neural network.

As shown in Figure 5, the RBF neural network contains three layers: an input layer, a hidden layer, and an output layer. The hidden layer is activated by an RBF. Let o and ω be the center and expansion constant of the RBF, respectively. Then, the main forms of the RBF can be expressed as

Formulas (18)–(20) are Gaussian function, inverse sigmoid function, and quasi-multi-quadratic function, respectively. Note that = ||a_t-o||; the smaller the value of ω, the narrower the function, and the higher the selectivity.

Training a multilabel learning model requires a lot of distinguishable information. To acquire more distinguishable information, the feature set containing the label attribute of each factor needs to be reconstructed, and the essential attributes of each label should be extracted from the original data on the corresponding teaching pressure factor. Fully considering the binary features of factor labels, this paper denotes the sample instances belonging to label b_l as and those not belonging to that label as X_l. Then, H_l can be defined as

X_l can be defined as

Next, k-means clustering is adopted to capture the attributes of H_l and X_l. H_l is divided into nonintersecting clusters with the cluster heads of {}, while X_l is divided into nonintersecting clusters with the cluster heads of {}. Among the multilabel learning samples of teaching pressure factors, |X_l| << |H_l| normally holds. This leads to the imbalance between positive and negative labels among the learning samples. To solve the problem, it is assumed that the number of positive labels equals that of negative labels:

Let η ∈ [0, 1] be the ratio parameter controlling the number of clusters. Then, the number of clusters of H_l and X_l can be configured by

The augmented feature sets and are imported to the neural network. The number of nodes on the hidden layer is set to 2f_l. The original vector of the basis function for label y_k is denoted as D_l = {}, i.e., the center of radial basis. Let DIS(A-D_l) be the Euclidean distance from the eigenvector to the original vector. By taking Gaussian function as the activation function, the activation function of the hidden layer can be expressed as

The formula of the expansion parameter Φ_j(k ≤ k ≤ y) can be rewritten as

Formula (26) shows that Φ_j is the mean distance between the original vectors of two basis functions. Thus, the mapping from the additional feature set of the original influencing factors to the feature set of labels can be determined, once the center and the expansion parameter of the radial basis are confirmed. Let be the true value of the i-th sample A_i on label b_k. Then, the mapping from a specific feature set to the output layer can be given bywhere V = [, , …, ] is the weight matrix calculated by the minimum quadratic sum of squares:

If = 1, A_i belongs to label b_l; if = 0, A_i does not belong to label b_l.

4. Experiments and Result Analysis

To facilitate subsequent modeling and data description, this paper defines nine independent variables for teaching pressure: overload pressure W₁, working duration pressure W₂, further education pressure W₃, title evaluation pressure W₄, pressure of changing teaching method W₅, pressure of conflict between personal life and work W₆, interpersonal interaction pressure W₇, pressure from personal quality defects W₈, and student management pressure W_9.

The nine teaching pressure factors were organized into a comparative series, and the quantified values B of teachers’ self-efficacy were grouped into a reference series. By the averaging method, the sample data on the teaching pressure and self-efficacy in a fixed period were nondimensionalized, to facilitate comparison and eliminate the influence of dimensionality. Table 1 shows the difference series between reference and comparative series. Further, the correlation coefficient and correlation degree between the two series were computed. The maximum and minimum absolute differences of the matrix were 0.513 and 0.003, respectively. Table 2 ranks the correlation degrees.

When the correlation degree is greater than 0.8, the teaching pressure factor is very highly correlated with self-efficacy; when the correlation degree is between 0.8 and 0.75, the two have a relatively high degree of correlation; when the correlation degree is between 0.75 and 0.7, the two have a general degree of correlation; when the correlation degree is between 0.7 and 0.65, the two are barely correlated; when the correlation degree is smaller than 0.65, the two are weakly correlated. As shown in Table 2, W₅, W₃, W₄, W₆, and W₁ are main influencing factors, while W₂, W₇, W₈, and W₉ are barely correlated factors.

The ridge variable selection model was constructed by ridge trace method and cross validation. Figure 6 presents the ridge traces of main influencing factors. From top to bottom, the five curves represent the regression curves of W₅, W₃, W₄, W₆, and W₁ generated with different values of the ridge regression parameter. As the parameter increased continuously from zero, the regression linearity of W₅ increased temporarily and then declined continuously, while the regression curves of W₃, W₄, W₆, and W₁ remained stable. The standard regression coefficient only oscillated very slightly.

The correlation degree between teaching pressure factors and teachers’ self-efficacy was predicted by the constructed model, with the data samples of 2014–2017 being the training set and those of 2018–2021 being the testing set. Tables 3 and 4 present the correlation analysis results based on multiple linear regression and RBF neural network, respectively. The two tables display the error, relative error, and MRE between the training set results and the test set results.

To verify its effectiveness, our model was compared with six other multilabel learning algorithms through simulation, namely, binary relevance, classifier chains, calibrated label ranking, random k-labelsets, machine learning–k-nearest neighbors (ML-KNN), and machine learning–decision tree (ML-DT). The performance was evaluated by five metrics: A: ratio of difference; B: error ratio of top-ranking label; C: mean distance between predicted label set and actual label set; D: error ratio of the ranking of error-prone labels; E: correct ratio of the ranking of high confidence labels. The simulation results of different models are displayed in Table 5. It can be seen that our model performed better than the 6 contrastive algorithms on all five metrics. Hence, the algorithm performance can be improved by fully utilizing the label correlations corresponding to the influencing factors.

Table 6 compares the runtime before and after pruning. Before the operation, the algorithm was highly complex. After the operation, the runtime was greatly shortened and was linearly correlated with the scale of the labels corresponding to the influencing factors. Because of the rising prediction accuracy of correlation degree, the proposed algorithm has an ideal overhead of time complexity.

5. Conclusions

This paper mainly analyzes and evaluates the relationship between teaching pressure and self-efficacy of college teachers based on artificial neural network. Firstly, the author created a GRA model for the teaching pressure and self-efficacy of college teachers and detailed the analysis procedure. Next, the possible multicollinearity of the GRA model was tested in two steps: computing the Pearson correlation coefficient and calculating the VIF. Then, a linear regression model was established based on Lasso variable selection model and ridge regression variable selection model and used to eliminate the multicollinearity between various teaching pressure factors in the GRA model. Finally, a multilabel learning algorithm was developed based on neural network and label correlation to automatically mine the correlations between the various teaching pressure factors and teachers’ self-efficacy.

Through experiments, the difference series between reference and comparative series, as well as correlation degree ranking, were obtained. The results show that overload pressure W₁, further education pressure W₃, title evaluation pressure W₄, pressure of changing teaching method W₅, and pressure of conflict between personal life and work W₆ are the main influencing factors, while working duration pressure W₂, interpersonal interaction pressure W₇, pressure from personal quality defects W₈, and student management pressure W₉ are barely correlated factors. After that, the ridge traces were plotted for the main influencing factors. The correlation analysis results were obtained based on multiple linear regression and RBF neural network. Moreover, the error, relative error, and MRE between the training set results and the test set results were displayed, and the runtime before and after pruning was summarized. The experimental results fully demonstrate the effectiveness of our model.

Data Availability

The data used to support the findings of this study are available from the author upon request.

Conflicts of Interest

The author declares no conflicts of interest.

References

N. A. Razak, H. Ab Jalil, and I. A. Ismail, “Challenges in ICT integration among Malaysian public primary education teachers: the roles of leaders and stakeholders,” International Journal of Emerging Technologies in Learning (iJET), vol. 14, no. 24, pp. 184–205, 2019.
View at: Publisher Site | Google Scholar
Y. Sun and F. Gao, “Exploring the roles of school leaders and teachers in a school-wide adoption of flipped classroom: school dynamics and institutional cultures,” British Journal of Educational Technology, vol. 50, no. 3, pp. 1241–1259, 2019.
View at: Publisher Site | Google Scholar
C. N. Chu, “Bring independent practice technology to after-school: preliminary results from the mixed roles of teachers and parents study to explore children effects in Chinese characters learning,” in Proceedings of the International Conference on Frontier Computing, pp. 841–845, Tokyo, Japan, June, 2016.
View at: Google Scholar
B. Wang, “Study on changes of college English teaching method and redefinition of teachers' roles in the computer network environment,” International Journal of Emerging Technologies in Learning, vol. 12, no. 8, pp. 137–145, 2017.
View at: Google Scholar
H.-L. Chou and J. C.-Y. Sun, “The moderating roles of gender and social norms on the relationship between protection motivation and risky online behavior among in-service teachers,” Computers & Education, vol. 112, pp. 83–96, 2017.
View at: Publisher Site | Google Scholar
O. Kwon, K. Choi, and M. Kim, “User acceptance of context-aware services: self-efficacy, user innovativeness and perceived sensitivity on contextual pressure,” Behaviour & Information Technology, vol. 26, no. 6, pp. 483–498, 2007.
View at: Publisher Site | Google Scholar
S. N. Hu and D. Chen, “Investigation on the present situation of university teachers’ scientific research pressure under the background of information,” in Proceedings of the 2020 International Conference On Artificial Intelligence And Education (ICAIE), pp. 249–252, Tianjin, China, June, 2020.
View at: Google Scholar
J. Pereira, A. Teixeira, and H. Simões, “Effect of Occupational Activity on Ambulatory Blood Pressure Behavior in university Teachers,” Occupational Safety and Hygiene V, vol. 2, pp. 563–566, 2017.
View at: Google Scholar
P. Dondon, J. M. Duprat, and C. Delesse, “Yoga practice: a modern tool against new kind of pressure for teachers and researchers in scientific engineering schools,” in Proceedings of the 7th WSEAS International Conference On Engineering Education, pp. 58–65, Corfu Island, Greece, July, 2010.
View at: Google Scholar
N. Canpolat, T. Pinarbasi, and M. Sözbilir, “Prospective teachers' misconceptions of vaporization and vapor pressure,” Journal of Chemical Education, vol. 83, no. 8, pp. 1237–1242, 2006.
View at: Publisher Site | Google Scholar
B. Standl and N. Schlomske-Bodenstein, “Investigating teacher’s empathy and its impact on pre-service teachers’ self-efficacy in programming tasks,” the 16th Workshop in Primary and Secondary Computing Education,” Virtual Event, Germany, vol. 1-2, 2021.
View at: Google Scholar
S. Latifah, N. E. Susilowati, K. Khoiriyah, and R. Rahayu, “Self-efficacy: its correlation to the scientific-literacy of prospective physics teacher,” Journal of Physics: Conference Series, vol. 1155, no. 1, Article ID 012015, 2019.
View at: Publisher Site | Google Scholar
A. A. Rogers, J. J. Rogers, and J. C. Baygents, “A longitudinal evaluation of an AP type, dual-enrollment introduction to engineering course: examining teacher effect on student self-efficacy and interest in engineering (evaluation),” in Proceedings of the 2019 ASEE Annual Conference & Exposition, Tampa, United States, June, 2019.
View at: Google Scholar
T.-Y. Chung and Y.-L. Chen, “Exchanging social support on online teacher groups: relation to teacher self-efficacy,” Telematics and Informatics, vol. 35, no. 5, pp. 1542–1552, 2018.
View at: Publisher Site | Google Scholar
P. Kuo, J. Huang, Q. Ji, L. Lin, Z. Chen, and K. Chen, “Comparison of self-efficacy, job stress and satisfaction of application-oriented university teachers in Fujian and Taiwan,” in Proceedings of the 2019 IEEE Eurasia Conference on IOT, Communication and Engineering (ECICE), pp. 109–112, Yunlin, Taiwan, October, 2019.
View at: Publisher Site | Google Scholar
M. A. Hamedi, “The relationship between educational psychology and academic motivation and self-efficacy and minor training skills of teachers,” Research Journal of Applied Sciences, Engineering and Technology, vol. 4, no. 23, pp. 5188–5192, 2012.
View at: Google Scholar
X. Wang and L. Wang, “The Reason and Self-Adjustment of College Teachers' Psychological Pressure under the Background of IT in Education,” in Proceedings of the 2011 IEEE International Symposium on IT in Medicine and Education, pp. 264–266, Cuangzhou, China, December, 2011.
View at: Google Scholar
Y. Xie, S. Zhang, and Y. Liu, “Abnormal behavior recognition in classroom pose estimation of college students based on spatiotemporal representation learning,” Traitement du Signal, vol. 38, no. 1, pp. 89–95, 2021.
View at: Publisher Site | Google Scholar
Y. Zhao, W. Ren, and Z. Li, “Prediction of English scores of college students based on multi-source data fusion and social behavior analysis,” Revue d'Intelligence Artificielle, vol. 34, no. 4, pp. 465–470, 2020.
View at: Publisher Site | Google Scholar
Q. Guo, “Detection of head raising rate of students in classroom based on head posture recognition,” Traitement du Signal, vol. 37, no. 5, pp. 823–830, 2020.
View at: Publisher Site | Google Scholar

Copyright

Copyright © 2022 Songjiao Wu. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

PDF Download Citation

Download other formats

Order printed copies

Views

154

Downloads

348

Citations