Research on cultivating students’ creative thinking ability in art design teaching based on machine learning

Creative thinking refers to the mode of thinking that has a new understanding of things, judgment, or can propose new solutions and new ways to solve problems. Creative thinking is of great significance to the cultivation of thinking ability [1-3]. Creative thinking has its own characteristics, such as dissimilarity, non-logic and compatibility. For example, dissimilarity is characterized by the use of unconventional methods of thinking and designing in the process of thinking and problem solving, and putting forward unique insights [4-7]. Non-logic refers to the process of thinking problems without thinking from a logical point of view step by step, can rely on the imagination or inspiration to think about the problem. Compatibility refers to the application of divergent thinking and fuzzy thinking in the process of problem solving, which can better ensure the effective and timely solution of problems [8-11].

Along with the deep implementation of the national education system reform process, the market under the new situation requires more and more high capacity and quality of talents. Art design is a very comprehensive discipline, which requires students to have interdisciplinary thinking and comprehensive ability [12-15]. Therefore, in the teaching of university art design courses teachers should not only focus on the training of theoretical knowledge ability, but also pay attention to the cultivation of students’ creative thinking, so as to better enhance the students’ dispersive thinking and ability [16-19], and then lay the foundation for the practice of university art design profession, etc., and machine learning, as an important field of artificial intelligence, helps to promote the teaching of art design students’ The cultivation of creative thinking ability [20-22].

As a highly practical discipline in colleges and universities, the art design program has a pivotal role in cultivating talents through practical teaching. Literature [23] emphasized the positive significance of art design courses in cultivating students’ design innovation ability, and put forward feasible strategies based on the cultivation of students’ innovation ability, aiming at cultivating artistic talents needed in the era. Literature [24] analyzes the impact of digital media art in art and design education and the problems that exist, points out the aging of the curriculum content, the confusion of the teaching mode and other status quo, and puts forward suggestions from the perspective of innovative thinking education. Literature [25] describes the content of professional integration and re-planning curriculum system based on the status quo of art professional talent cultivation, and focuses on the analysis of the cultivation strategy of applied innovative talent courses, aiming to improve the level of China’s modernization. Literature [26] puts forward the strategy of establishing a master’s innovation training mode based on cdio in view of the challenges faced by the master’s training of art and design. Through the construction and implementation of the “trinity” innovation training system, the effectiveness of this system is emphasized, and it has certain reference value for the innovation strategy of art and design professional talent cultivation. Literature [27] discusses the talent cultivation strategy of art and design “craftsmanship”, aiming to provide a reference for the transformation of China’s art and design curriculum to adapt to national needs. Based on the current situation and challenges of the art design profession, literature [28] puts forward the proposal of constructing a “double-body” education system and establishing a “multiple” evaluation system, emphasizing the important role of the modern apprenticeship talent cultivation mode in the art design profession. The above research reveals the importance of art and design profession for talent cultivation, while there are relatively few studies on “cultivating students’ creative thinking ability in art and design teaching”, which indicates that the cultivation of creative thinking ability in art and design has not been paid attention to by academics.

In today’s society, creative thinking has become a necessary ability for students, and cultivating students’ creative thinking ability in teaching not only helps students’ personal development, but also provides strong support for their future career development. Literature [29] explored the cultivation of students’ creative thinking in the teaching of functions of complex variables and sought different teaching strategies and methods to stimulate students’ interest, so as to improve their innovative and practical abilities and promote their overall development. Literature [30] and literature [31] elaborated on the cultivation of students’ creative thinking ability through art education, and emphasized the importance of cultivation strategies based on the deficiencies in the cultivation of creative thinking ability by art education, with a view to better serving the development of society. Literature [32] discusses the cultivation of students’ creative thinking ability in piano curriculum education, reveals the problems existing in the current piano curriculum education, and puts forward effective strategies. Literature [33] illustrated that teachers’ optimization of teaching content and holding art competitions not only helped to improve the efficiency and quality of teaching, but also helped to improve students’ creative thinking ability. Literature [34] emphasized the application of artificial intelligence in education. It expresses that artificial intelligence is not only conducive to improving students’ independent thinking and critical art appreciation, but also has a positive impact on cultivating students’ creative thinking ability. In summary, scholars agree that creative thinking plays an important role for students, and that in the field of education, art education, piano education, and information education that integrates disciplines have a positive impact on cultivating students’ creative thinking, but there are relatively few studies on the cultivation of creative thinking based on machine learning.

In order to explore more deeply the influence factors of cultivating students’ creative thinking ability in art design teaching, so as to provide improvement suggestions for teaching. This paper collects and analyzes literature related to creative thinking, and initially constructs a set of creative thinking evaluation index system based on art design teaching. Invite scientific career experts to use the Delphi method to conduct three rounds of research on the indicators of the system, optimize the indicators based on the results of the research, and further strengthen the scientific nature of the system. At the same time, in order to ensure the objective assessment of students’ performance under the index system in this paper, this paper uses BP neural network to simulate the algorithm of the evaluation system. And put into the actual assessment. Based on the evaluation results and analysis of the algorithmic simulation, this paper puts forward a number of suggestions for the cultivation mode of students’ creative thinking ability.

2

Construction and testing of an evaluation index system for creative thinking

2.1

Preliminary construction of the evaluation indicator system

In the preliminary construction of evaluation indexes, the literature method is mainly used to summarize and analyze from a large number of previous studies, which have been reviewed earlier. By analyzing and studying the existing results of creative thinking evaluation, the evaluation indexes of students’ creative thinking cultivated in art and design teaching based on machine learning are initially constructed, as shown in Figure 1. Among them, there are 3 first-level indicators, which are creative thinking tendency (A), creative thinking ability (B), and creative thinking monitoring (C). There are 12 second-level indicators, which are curiosity (A1), imagination (A2), challenging (A3), risk-taking (A4), persistence (A5), independence and cooperation consciousness (A6); asking new questions (B1), presenting new evidence (B2), making new interpretations (B3), drawing new conclusions (B4); self-monitoring (C1), self-regulation (C2).

2.2

Analysis of the results of the three rounds of the Delphi method

A total of 40 scientific experts were invited to participate in the Delphi questionnaire. The 40 experts included: 20 scientists, 13 science educators, and 7 philosophers of science.

2.2.1

First round of Delphi survey research

The questionnaire for the first round of the Delphi study was distributed to 40 experts on February 12, 2022 and was returned in full on March 16th. The return rate of the questionnaire was 100%. The objective questions were statistically analyzed as shown in Table 1. Through the analysis, it was found that the experts’ average scores for each indicator exceeded 3.70, and the plurality of the remaining indicators was 5 except for the plurality of the persistence indicator (A5) and the plurality of the self-regulation (C2) indicator, indicating that the experts had reached a consensus on the 10 indicators of creative thinking; among them, the average score for the ability to raise new questions (B1) in the ability to think creatively (B) was the highest at 4.95, followed by the average score for the creative thinking Curiosity (A1) and Imagination (A2) in Tendency (A), with a mean of 4.91, and the mean scores of these three indicators exceeded 4.90, indicating that the expert group generally agreed that creative thinking of middle school students based on the cultivation of top-notch innovators requires posing new questions (B1), curiosity (A1) and imagination (A2). Experts’ comments on the subjective questions corresponding to the “Evaluation indicators of creative thinking of secondary school students based on the cultivation of top-notch innovative talents” were fewer, and some of the experts thought that the connotation of the evaluation indicators provided was unclear, so the connotation of the evaluation indicators was revised based on the feedback from experts. A second round of Delphi questionnaires and a feedback report from the first round of the Delphi study were generated by revising the first round questionnaires.

Table 1.

Results of objective question data analysis in the first round of Delphi survey

Productive thinking		Average value	Smple mode	Standard deviation
A	A1	4.91	5	0.35
	A2	4.91	5	0.48
	A3	4.86	5	0.38
	A4	4.53	5	0.82
	A5	4.89	4	0.33
	A6	4.80	5	0.50
B	B1	4.95	5	0.14
	B2	4.79	5	0.55
	B3	4.75	5	0.47
	B4	4.82	5	0.43
C	C1	4.53	5	0.53
C	C2	4.47	4	0.48

2.2.2

Second round of Delphi survey research

On April 4, 2022, we sent the second round questionnaire to 40 experts who participated in the first round of the Delphi survey study.The questionnaire collection was finished on April 30, and out of the 40 experts, a total of 37 experts participated in the second round of the study, among which 3 experts did not participate in the second round of the study, and the questionnaire collection rate was 92.5%. The data processing of the second round was the same as the first round, and the data statistics on the scores of the objective questions were carried out, as shown in Table 2. The results showed that the experts had a high degree of recognition of the division of the 12 indicators, and the average score of each indicator was over 3.71, and the plural was 5.

Table 2.

Objective question data analysis of the second round of Delphi survey

Productive thinking		Average value	Sample mode	Standard deviation
A	A1	4.59	5	0.60
	A2	4.75	5	0.66
	A3	4.65	5	0.63
	A4	4.70	5	0.58
	A5	4.45	5	0.73
	A6	4.65	5	0.76
B	B1	4.61	5	0.45
	B2	4.72	5	0.65
	B3	4.84	5	0.44
	B4	4.80	5	0.47
C	C1	4.66	5	0.71
C	C2	4.72	5	0.61

2.2.3

Third round of Delphi method survey research

On May 5, 2022, we sent the third round questionnaire to 37 experts who participated in the second round of the Delphi study, and the questionnaire collection ended on May 27, with 35 recoveries and a questionnaire recovery rate of 94.60%. Table 3 lists the relevant data of the second and third round questionnaires and their processing results. From Table 3, it can be clearly seen that the average scores of the 12 indicators of creative thinking are all above 3.66, that is, the expert consensus degree of each indicator meets the standard, and the stability of all 12 indicators is greater than 66.7%, which meets the standard of stability. As can be seen from the average scores of each indicator in the third round of the Delphi study, the highest average score was for Asking New Questions (B1), which was as high as 4.90; this indicator also scored the highest in the first round of the Delphi study.

Table 3.

Creative thinking second, third round of objective question data analysis

Index	Ihe second round			Third round			Stability(%)
Index	Average value	Sample mode	Standard deviation	Average value	Sample mode	Standard deviation	Stability(%)
A1	4.59	5	0.60	4.88	5	0.42	90.30
A2	4.75	5	0.66	4.76	5	0.66	82.99
A3	4.65	5	0.63	4.80	5	0.41	76.59
A4	4.70	5	0.58	4.53	5	0.77	76.58
A5	4.45	5	0.73	4.72	5	0.56	74.23
A6	4.65	5	0.76	4.72	5	0.58	86.52
B1	4.61	5	0.45	4.93	5	0.33	91.56
B2	4.72	5	0.65	4.82	5	0.45	82.77
B3	4.84	5	0.44	4.77	5	0.63	74.69
B4	4.80	5	0.47	4.72	5	0.53	66.88
C1	4.66	5	0.71	4.72	5	0.51	73.15
C2	4.72	5	0.61	4.73	5	0.44	76.79

2.3

Effectiveness test of the optimized evaluation index system

In order to test the effectiveness of the optimized evaluation index system established in this paper in practice, this study invites four experts to evaluate the classroom performance of three students X, Y, and Z in the same group in the art teaching classroom based on the optimized evaluation index system established in this paper (referred to as Evaluation Indicator System I) and the adjusted evaluation system of secondary school students’ innovative ability proposed in previous studies (referred to as Evaluation Indicator System II) respectively. Two evaluations.

2.3.1

Implementation of the validity test of the evaluation indicator system

In order to facilitate the comparative analysis of data, based on the different evaluation system scores need to be converted into a percentage system, because the two evaluation index system contains 12 secondary indicators, so you can set the full score of each indicator for 10 points, the total score is 120 points. The scores for the three dimensions of creative thinking tendency, creative thinking ability and creative thinking monitoring under the evaluation index system I are 60, 40 and 20 respectively; the scores for the dimension of idea generation (D) under the evaluation index system II are 50, containing five secondary indexes of flexibility (D1), fluency (D2), originality (D13), imaginativeness (D4), and intuition (D5). The Creative Practice (E) dimension is worth 40 points and contains 4 Level 2 indicators, namely Knowledge Base (E1), Problem Identification (E2), Design Solution (E3), and Practical Solution (E4); and the Analytical Reasoning (F) dimension is worth 30 points and contains 3 Level 2 indicators, namely Interpretation (F1), Evaluation (F2), and Inference (F3). Correspondingly, the score standard of each evaluation index is divided into 5 levels, of which 1~2 points represent “very poor” student performance, 3~4 points represent “poor” student performance, 5~6 points represent “average” student performance, 7~8 points represent “good” student performance, and 9~10 points represent “good” and “excellent” student performance. According to the performance level of the students in the classroom, the experts scored the indicators as appropriate, and finally the average and standard deviation of the scores of each student by the four experts. The two evaluations of the three students by experts according to different index systems are shown in Table 4 and Table 5.

Table 4.

Student evaluation index system I score table

Primary index	Secondary index	Average score			Score standard deviation
Primary index	Secondary index	Student X	Student Y	Student Z	Student X	Student Y	Student Z
A	A1	7.30	6.71	7.48	0.51	0.51	0.55
	A2	8.60	5.51	7.23	0.57	0.59	0.57
	A3	8.58	5.26	7.76	0.56	0.50	0.53
	A4	8.30	6.24	6.74	0.95	0.53	0.95
	A5	8.29	6.25	7.00	0.52	0.52	0.84
	A6	9.00	7.03	7.54	0.78	0.23	0.56
	Dimensional total score	50.07	37.00	43.75	2.26	0.80	2.11
B	B1	9.28	4.76	8.26	0.51	0.51	0.51
	B2	8.27	6.23	8.51	0.52	0.97	0.57
	B3	8.53	4.24	8.74	0.57	0.53	0.79
	B4	8.15	3.70	8.01	0.50	0.50	0.83
	Dimensional total score	34.23	18.93	33.52	0.58	0.49	0.51
C	C1	8.58	4.76	8.51	0.57	0.50	0.47
	C2	6.30	6.23	6.56	0.49	0.55	0.57
	Dimensional total score	14.88	10.99	15.07	1.56	0.55	1.13
	total points	99.18	66.92	92.34	2.84	1.33	3.34

Table 5.

Student evaluation Index System II score sheet

Primary index	Secondary index	Average score			Score standard deviation
Primary index	Secondary index	Student X	Student Y	Student Z	Student X	Student Y	Student Z
D	D1	6.47	2.98	5.21	0.58	0.84	0.96
	D2	7.51	4.23	6.22	0.57	0.48	0.53
	D3	7.02	4.01	6.74	0.79	0.81	0.96
	D4	6.71	4.02	5.02	0.94	0.82	0.84
	D5	6.23	4.76	6.11	0.53	0.97	0.81
	Dimensional total score	33.94	20.00	29.3	2.41	1.84	2.03
E	E1	6.23	4.21	5.01	0.51	0.52	0.81
	E2	7.03	6.03	5.21	0.84	0.83	0.94
	E3	7.24	4.12	3.74	0.50	0.82	0.53
	E4	6.73	4.23	4.52	0.51	0.53	0.61
	Dimensional total score	27.23	18.59	18.48	1.28	1.74	1.32
F	F1	7.03	6.21	4.23	0.81	0.52	0.83
	F2	7.45	4.74	4.28	0.52	0.50	0.52
	F3	6.56	4.75	5.03	0.58	0.51	0.80
	Dimensional total score	21.04	15.70	13.54	2.33	1.24	2.57
	total points	82.21	54.29	61.32	3.79	4.14	4.09

2.3.2

Analysis of the validity test of the evaluation index system

A comprehensive comparison of the overall ratings of the three panelists by the above four experts based on the two types of evaluation indicator systems shows that: for student X, the mean value of the expert group’s evaluation scores based on evaluation indicator system I was 99.18, with a standard deviation of 2.84, while the mean value of the expert group’s evaluation scores based on evaluation indicator system II was 82.21, with a standard deviation of 3.79; for student Y, the mean value of the expert group’s evaluation scores based on evaluation indicator system I was 66.92 with a standard deviation of 1.33, while the mean of the evaluation scores of the expert group based on the evaluation index system II was 92.34 with a standard deviation of 4.14; for Student Z, the mean of the evaluation scores of the expert group based on the evaluation index system I was 82.47 with a standard deviation of 3.31, while the mean of the evaluation scores of the expert group based on the evaluation index system II was 61.32 with a standard deviation of 4.09. The standard deviation is 4.09.

It can be seen that for the same student, the application of evaluation index system I has a higher mean and lower standard deviation than evaluation index system II; comparing students X, Y and Z, it can be found that student Y scores lower under both evaluation index systems, which is also consistent with the evaluation of classroom instructors, according to the classroom instructors’ description of student Y’s degree of participation and concentration in the process of in-group project exploration. Weak. In addition, according to the evaluation index system I, student A has the best classroom performance, while according to the evaluation index system II student Z is more prominent, according to the description of the classroom instructor, student C serves as the group leader, according to the long-term teaching observation of the classroom instructor, the overall performance of student C is better than the other members in all aspects, and the performance of student X is easy to fail to grasp the key points in the inquiry session, although the performance of student X is positive and active. Therefore, in combination with the evaluation of on-site lecturers, it can be found that the result indicates that the evaluation index system constructed in this study is more reasonable for the evaluation system of innovation ability of secondary school students selected by the study, and its application is better than the evaluation indexes currently available.

3

Creative thinking evaluation model

Humans have always been able to evaluate and react to many non-linear problems, and have an efficiency in solving certain complex problems that cannot be matched by machines. The basis of this lies in the biological neural network composed of countless interconnected neurons in the brain, in which neurons interact with each other through bioelectrical current stimulation and react to each other to get results. The researchers were inspired by this, using computer algorithms to simulate the biological neural network in the brain, mimicking the brain’s work process, so that the computer has a similar way of thinking to human beings, in order to achieve the ability to solve certain non-linear problems.

3.1

Training process of BP neural network

Take the following three-layer BP neural network as an example to briefly introduce the training process of BP neural network. Figure 2 shows the topology of a three-layer BP neural network.

As shown in the figure, let the input neuron node be $x_{i}$ , the hidden layer neuron node be $y_{j}$ , and the output neuron node be $z_{l}$ . The connection weight between the input neuron node and the hidden layer neuron node is $w_{j i}$ , and the connection weight between the hidden layer neuron node and the output layer neuron node is $v_{l j}$ . The threshold for each neuron node of the hidden layer is $θ_{j}$ , and the threshold for each neuron node of the output layer is $θ_{l}$ , and the excitation function is $f$ , and the desired output of the output neuron node is $t_{l}$ .

The output of each neuron node in the hidden layer is equation (1): (1) $y_{j} = f (\sum w_{j i} x_{i} - θ_{j}) = f (n e t_{j})$

The output of each neuron node in the output layer is equation (2): (2) $z_{l} = f (\sum v_{l j} y_{j} - θ_{l}) = f (n e t_{l})$

The error function between the actual output $z_{l}$ and the desired output $t_{l}$ is equation (3): (3) $E = \frac{1}{2} \sum_{l} {(t_{l} - z_{l})}^{2}$

$E$ Derive the connection weights between the neuron nodes in the hidden layer and the neuron nodes in the output layer as equation (4): (4) $\frac{\partial E}{\partial v_{l j}} = - (t_{l} - z_{l}) \times f' (n e t_{l}) \times y_{j} = - δ_{l} y_{j}$

Then the error of each neuron node in the output layer is equation (5): (5) $δ_{l} = (t_{l} - z_{l}) \times f' (n e t_{l})$

$E$ The connection weights between each neuron in the output layer and each neuron in the hidden layer are derived as equation (6): (6) $\frac{\partial E}{\partial w_{j i}} = - \sum_{l} δ_{l} v_{l j} f' (n e t_{j}) \times x_{i} = - δ_{j} x_{i}$

Then the error of each neuron node in the hidden layer is equation (7): (7) $δ_{j} = f' (n e t_{j}) \times \sum_{l} δ_{l} v_{l j}$

The connection weights between each neuron in the hidden layer and each neuron in the output layer are varied as equation (8): (8) $Δ v_{l j} = δ_{l} = η δ_{l} y_{j}$

The connection weights between each neuron in the input layer and each neuron in the hidden layer are varied as equation (9): (9) $Δ w_{j i} = δ_{j} = η' δ_{l} x_{i}$

$E$ Thresholding for each neuron in the output layer is derived as equation (10): (10) $\frac{\partial E}{\partial θ_{l}} = (t_{l} - z_{l}) \times f' (n e t_{l}) = δ_{l}$

The threshold changes as equation (11): (11) $Δ θ_{l} = η \frac{\partial E}{\partial θ_{l}} = η δ_{l}$

$E$ Thresholding for each neuron in the hidden layer is derived as equation (12): (12) $\frac{\partial E}{\partial θ_{j}} = \sum_{l} (t_{l} - z_{l}) \times f' ({net}_{l}) \times v_{l j} \times f' ({net}_{j}) = δ_{j}$

The threshold changes as equation (13): (13) $Δ θ_{j} = η' \frac{\partial E}{\partial θ_{j}} = η' δ_{j}$

Figure 3 shows the flowchart of the BP neural network algorithm:

3.2

Genetic algorithms

If an artificial neural network is trained to operate by simulating the biological neural network of individual organisms in nature, then a genetic algorithm is trained by simulating the iterative evolutionary process of biological populations in nature. Genetic algorithm is a stochastic search optimization method that evaluates the quality of the solution by the fitness. Currently genetic algorithms are widely used in artificial intelligence and other fields, as optimization algorithms have a broad application prospect.

The core idea of genetic algorithm is the principle of natural selection in nature, that is, the principle of the survival of the fittest. Genetic algorithm can effectively avoid falling into some traditional algorithms of the solution error. The solution of genetic algorithm can be updated and iterated by mutation and selection, so that the solution is closer to the global optimal solution, thus achieving local improvement and global optimization.

Genetic algorithms follow the laws of selection and survival of the fittest. Firstly, an initial solution is input from the required problem, which is transformed into binary code, that is, genes and genome, so as to simulate the way of evolution of biological populations in nature, and through mutation, selection, etc., the populations will gradually evolve, and finally get an individual of the highest quality, that is, the optimal solution of the required problem. Each iteration of the population becomes a generation, and then the fitness function is calculated for each individual, eliminating low-quality individuals, and then using the theory of genetic variation in genetics, new individuals are constructed, and the new individuals form a new generation, which will continue to carry out the calculation and iteration of the fitness function until the value of the fitness meets the target requirements, and the final generation is called the most excellent generation, which is the optimal solution of the problem being solved, at which point the algorithm stops.

The three kinds of turnover of genetic algorithm are selection, crossover, and mutation, and the excellent individuals are retained in the iteration to enter the next generation of population, and after many iterations of evolution, the fitness is effectively improved to complete the search for optimal solution, genetic algorithm has many advantages, as follows: 1)

Can perform parallel computation. Genetic algorithm does not focus on a single individual, but focuses on all the individuals in the entire population, in this way, the search efficiency of the genetic algorithm will be higher, and it is in the global search. At the same time, the search can also constantly adjust the direction, which further improves the search efficiency.

2)

It is not easy to fall into local optimization. Most of the traditional search methods are single-point search, so it is very easy to fall into local optimization when solving multi-polar problems, while the object of the genetic algorithm is the entire population, which belongs to the group search, and can well avoid this problem.

3)

Good expandability. As an optimization algorithm, genetic algorithm can be well combined with other algorithms, and can optimize other algorithms such as BP neural network, which has excellent expandability.

3.3

Optimization of BP neural network weights and broad values by genetic algorithm

The basic principle of genetic algorithm to optimize the weights and thresholds of BP neural network is: using the characteristics of chromosomes in genetic algorithm, encoding the weights and thresholds of BP neural network, using certain methods to generate the initial population, using the cost function of the fitting error of BP neural network as the fitness function of genetic algorithm, and through the genetic operation, screening the individuals and selecting the best chromosomes as the BP neural network’s weights and thresholds.

3.3.1

Coding and Initial Population Generation

Encoding the connection rights to the network. Binary encoding and floating point encoding are the most commonly used encoding methods. Among them, binary encoding is simple to operate, but not intuitive and not high in precision; floating point encoding is very intuitive and high in precision. Since the connection weights of the BP neural network for safety evaluation of university laboratories are floating-point numbers, floating-point number coding method is chosen here to encode the connection weights. That is, the weights and queues of the BP neural network are cascaded in a certain order to form an array of floating-point numbers, which is used as a chromosome of the genetic algorithm.

The safety evaluation model of university laboratory adopts three-layer BP network structure, and the number of nodes in the input layer, implicit layer and output layer are set to be $N$ , $S$ and $M$ , respectively, and the length of the coding is given by Eq. (14) (14) $R = N \times S + S \times M + S + M$

$X$ chromosomes of length $R$ were randomly generated, i.e., the initial population was formed.

Since the topology of this network is 24-9-1, the coding length is $R = 24 \times 9 + 9 \times 1 + 9 + 1 = 235$ .

For the determination of the number of populations $Y$ , too large will lead to slow convergence of the network, while too small will reduce the accuracy of network training. Therefore, in this paper, the value of $Y$ is 30.

3.3.2

Determination of the fitness function

The $R$ connection weights in the initialized population are given to the BP network for forward propagation of the input signal and the sum of squares of the errors between the output value of the network and the desired value is calculated $E (i)$ , and the fitness function is set to be Eq. (15): (15) $f (i) = \frac{1}{E (i)} = \frac{1}{\sum_{k = 1}^{n} {(y_{k} - a_{k})}^{2}}$

Combining the genetic algorithm with the evaluation criteria of the BP network, the smaller the sum of squares of errors, the higher the adaptation, i.e., the better the network performance.

3.3.3

Genetic manipulation

1)

Selection

Adaptation-based sorting allocation method is used for selection, that is, first calculate the fitness of each individual in the population, and then sort all the individuals in the population according to the size of their fitness, and the probability of each individual to be selected is allocated with sorting results, and the principle of allocation is that a large fitness value corresponds to a high probability of selection, and a small fitness value corresponds to a low probability of selection.

The advantage of the selection method of assigning the proportion according to the order of fitness is that the individuals with high fitness can be selected as the individuals to be inherited to the next generation of the population, and at the same time, it can effectively avoid the problem of early maturity of the super individuals in the population which is easy to be generated by using the roulette wheel to select the individuals.

2)

Crossover

The most important operation in the genetic algorithm is the crossover operator, where new individuals are generated by the population through crossover, continuously expanding the search space, and ultimately achieving the purpose of global search. Crossover is the crossover of the gene chains of two selected individuals with a certain probability to generate two new individuals.

Since this paper adopts floating-point encoding, the crossover operator adopts the arithmetic crossover method of floating-point type. Assuming that $x_{1}$ and $x_{2}$ in the population are individuals of the father, the offspring ${x^{'}}_{1}$ and ${x^{'}}_{2}$ produced by both parents of the father are given in equation (16): (16) $\{\begin{array}{l} {x^{'}}_{1} = a x_{1} + (1 - a) x_{2} \\ {x^{'}}_{2} = a x_{2} + (1 - a) x_{1} \end{array}$ where $a \in (0, 1)$ .

The value of the crossover probability is generally taken between 0.4 and 0.99, and in this paper the crossover probability is taken to be 0.8.

3)

Mutation

In order to maintain the diversity of the population, the variation operator is used to generate new individuals. Mutation randomly changes the value of a string in the string structure data with a certain probability for the selected individuals.

In this paper, non-uniform mutation is used to make a random perturbation of the original gene values and the new gene values after mutation use the result of that perturbation. A small amount of adjustment is made to all the gene loci at a time with equal probability. The value of the mutation probability is taken between 0.001 and 0.1.

3.3.4

Generation of new populations

After genetic operation on the original individuals with crossover and mutation operators, new individuals are generated, and the new individuals are inserted into the original population to generate a new population. Calculate the fitness value of the new individual to determine whether the number of cycles or optimization criteria is reached, if so, proceed to the next step, otherwise continue to cycle through the genetic operation.

3.3.5

Generation of initial weights for BP neural networks

After the genetic algorithm reaches the maximum number of genetic generations or the set index, the optimized network connection weights are the decoding values of the optimal individuals in the final group.

After the above research design, the genetic algorithm is combined with BP neural network to construct an algorithmic simulation model of creative thinking evaluation model based on BP neural network, which is able to more objectively assess the students’ performance under the index system of this paper, and it will be adopted in further practical assessment.

4

Actual assessment results of the algorithmic simulation model

In order to make the selected samples representative, 10 students from a comprehensive university with different majors and grades were chosen for this assessment, and the students’ professional backgrounds involved science, arts, and medicine; the grade distribution included undergraduate and graduate students. Table 6 shows the composite scores of students of different majors and grades in this assessment.

Table 6.

S₁~S₁₀ Students creative thinking ability total score

Evaluation object	Major/Grade	Creative thinking ability score
S₃	Law/Freshman	80.32
S₈	Art/Freshman	71.26
S₄	Chemical Engineering/Sophomore	84.21
S₁₀	Music/Sophomore	74.55
S₁	Life Sciences/Junior year	89.92
S₆	Computer and Information Technology/Junior year	91.20
S₅	Foreign language/Senior	84.87
S₉	Physics and Electronic Technology/Senior year	89.00
S₂	Education/Postgraduate	83.16
S₇	Management/Postgraduate	84.79

As can be seen from Table 6, in general, the status of information literacy skills of most students in Chinese universities is still in the middle at present, and only a few students have relatively good information literacy skills. The main reasons are as follows: 1)

Different majors: Although most students have received information literacy education before entering colleges and universities, due to the different regions, teaching levels and degrees of importance, the results of the survey show that there are great differences between students of different majors. Students in science and engineering are better than students in liberal arts and arts in terms of information literacy skills.

2)

Different grades: the length of time spent in college and university directly affects students’ information literacy skills. The longer they have been in school, the more opportunities they have to receive information literacy education, the more they are influenced by the environment, and the more their own knowledge structure system will be gradually improved. Therefore, Table 6 can clearly show that the information literacy ability of graduate students and juniors and seniors is better than that of freshmen and sophomores.

3)

Analysis of specific individual abilities. Most of the students use information technology in order to obtain information, the existing ability of the students is still relatively strong, but there is still a lot of room for development and improvement in the evaluation and innovation links.

From the results of the above analysis, it can be seen that at this stage, the problem in front of teachers of various disciplines in colleges and universities and librarians in colleges and universities is that they must vigorously improve the mode of teaching and service, stimulate the interest of students in information acquisition, regularly provide students with theoretical and practical opportunities, strengthen the practical ability of students, the ability to conduct independent inquiries, and the ability to continuously develop innovative knowledge, etc., so as to comprehensively improve the students’ ability to improve their information literacy.

5

Suggestions for improving the teaching of students’ creative thinking skills

Based on the construction and optimization of the evaluation index system of creative thinking in this paper, combined with the results of the actual test of the evaluation index system of this paper, this paper puts forward the following three improvement suggestions for the cultivation of students’ creative thinking ability in art teaching.

5.1

Changing the old teaching model and reflecting the students’ subjective position

In the old education model, subject to strict education management teachers often do not tend to let students question the authority and reflect on the answer, which leads to students in the learning process, almost only looking for the only standard answer to the question, and in the mind to form a stereotype of the thinking of the answer to be identified. This model of education has stifled too much imagination and strangled the germ of creative thinking in the cradle. Naturally, when students study art and design, a profession that highly requires creative thinking, they need to spend great efforts to make up for this deficiency. Therefore, in the actual teaching process of higher vocational college art and design majors, teachers should actively create opportunities for students to question questions and provide necessary help for students to find scientific and reasonable answers in multiple directions. In the teaching process, to create a learning environment that dares to reflect on the status quo, question authority, think on their own, and think diffusively, without being bound by the restrictions of routine and common sense. In addition, teachers should not only teach students the relevant knowledge and skills of art and design, but also encourage and guide students when they face the problem of reverse thinking. In the teaching process of higher vocational college art design majors, the students’ main position in the learning process should be reflected. Teachers should guide students to independent learning and stimulate the progress of students’ comprehensive ability.

5.2

Promote the development of students’ creative consciousness in various aspects

In the specific teaching process, should be as far as possible in each course to set up open to guide students to divergent thinking links, in letting students steadily learn art design professional knowledge at the same time, comprehensively give full play to their personal artistic imagination, stimulate the sense of innovation. Students in the learning process, to learn to summarize the knowledge, from all aspects of art design to think, and finally summarize their own feelings and views.

5.3

Putting it to the test and moving forward with head held high

In the process of learning art and design majors, the main way for students to acquire knowledge and information is through the teacher’s teaching in the classroom. However, the formation and improvement of students’ creative thinking ability need to be accumulated little by little in the actual design practice. Although art is often higher than life, the value of art design is to meet the needs of life. Therefore, in order to let students grow up and mature in the practice of life, and form their own unique observation perspective and insight, teachers also need to let students actively participate in the actual art design projects, forcing students to change from passive thinking to independent thinking in the actual hands-on process, and gradually cultivating students’ creative thinking ability.

6

Conclusion

In the preliminary construction and optimization of the creative thinking evaluation index system, this paper finds that: taking the initiative to ask new questions, i.e., actively thinking about problems and exploring things, is the key to the development of students’ creative thinking ability. In the round of evaluation after the algorithmic simulation of the creative thinking evaluation model, this paper found that, rather than receiving theoretical knowledge in the classroom, it is more likely to promote the growth of creative thinking ability by encouraging students to practice theories in practice and explore innovation in practice. Based on the results of the two rounds of research in this paper, we propose (1) changing the subject: focusing on students in the classroom, encouraging them to question and guiding them to think, and (2) actively participating: accumulating experience and thinking independently in practice.

This paper focuses on the performance dimensions and cultivation methods of creative thinking ability, invites scientific career experts to test the performance dimensions and modify them, and introduces BP neural network and genetic algorithm to optimize them. The analysis of the actual assessment results guided the proposal of the improvement of the cultivation method, which provides a strong reference for the cultivation of students’ creative thinking ability.

Language:: English

Publication timeframe:: 1 times per year
Journal Subjects:: Life Sciences, Life Sciences, other, Mathematics, Applied Mathematics, General Mathematics, Physics, Physics, other

Journal RSS Feed

Research on cultivating students’ creative thinking ability in art design teaching based on machine learning

Min Li

Hongying Song

Published Online: Sep 23, 2025

Received: Feb 05, 2025

Accepted: May 11, 2025

DOI: https://doi.org/10.2478/amns-2025-0990

KeywordsCreative thinking ability, Evaluation index system, BP neural network, Genetic algorithm

© 2025 Min Li and Hongying Song published by Sciendo

This work is licensed under the Creative Commons Attribution 4.0 International License.

Keywords
Creative thinking ability, Evaluation index system, BP neural network, Genetic algorithm