Analysis of entrepreneurship education in colleges and based on improved decision tree algorithm and fuzzy mathematics

Abstract

In order to improve the performance of entrepreneurship and innovation education in colleges and universities, this study attempts to build an evaluation system and model of innovation and entrepreneurship in colleges and universities to provide a complete and practical tool for government education authorities and universities to evaluate the implementation of innovation and entrepreneurship education. In this research, decision tree and fuzzy mathematics are used as the basis of the model algorithm, and the algorithm is improved based on the analysis of traditional algorithms. Moreover, based on the improved decision tree algorithm, an evaluation index system for university innovation and entrepreneurship education is constructed. After determining the evaluation indicators of innovation and entrepreneurship education in colleges and universities, this study uses several universities as examples to analyze and define the definitions of various indicators. In addition, this study statistically analyzes the results of entrepreneurship and innovation education in colleges and universities through simulation. The research shows that the model proposed in this paper has a certain practical effect, and based on the simulation results, this study makes several suggestions.

Keywords

Neural network improved algorithm path ranking network education knowledge recommendation

1 Introduction

Since the 17th National Congress of the Communist Party of China, education and other related administrative departments have successively released relevant policies and measures on “promoting employment through entrepreneurship” and encouraged the implementation of innovative entrepreneurship education in universities across the country. In 2014, Premier Li Keqiang of the State Council issued a call for “mass entrepreneurship and innovation” at the Davos Forum and included it in the 2015 government work report. In June 2015 and July 2017, the State Council successively put forward “Opinions on Vigorously Promoting Popular Entrepreneurship and Innovation of Several Public Policies” and “Opinions on Strengthening the Implementation of Innovation-Driven Development Strategies and Further Promoting Mass Innovation and Public Innovation and Further Development”, which once again pushed innovation and entrepreneurship education to a climax [1]. In 2018, the State Council put forward “Opinions on Promoting the High-quality Development of Innovation and Entrepreneurship and Creating an Upgraded Version of “Double Innovation””, which emphasized the important role of innovation and entrepreneurship education in promoting economic growth. In 2019, the State Council’s “National Vocational Education Reform Implementation Plan” proposed that “establish and improve school settings, teaching staff, teaching materials, information construction, safety facilities and other school-running standards to lead the development of vocational education services and promote employment and entrepreneurship.” It can be seen that the education of innovation and entrepreneurship in colleges and universities is an important task for a long time. However, compared with developed countries, the innovation and entrepreneurship education in China’s universities is still in its infancy. To promote innovation and entrepreneurship education in colleges and universities, we must accelerate the establishment of an evaluation model and evaluation system for the quality of innovation and entrepreneurship [2].

The education of innovation and entrepreneurship is aimed at cultivating talents with the basic qualities of entrepreneurship and pioneering personality, so that students can establish an awareness of entrepreneurship and form an ability to innovate and entrepreneurship. In the past, innovation and entrepreneurship education was mainly implemented for the society. After the Party’s “17th National Congress” report put forward the concept of “promoting employment through entrepreneurship”, innovation and entrepreneurship education began to be implemented for all school students. After the State Council ’s “Government Work Report” proposed the strategic deployment of “mass entrepreneurship and innovation” in 2015, schools at all levels and various levels set off a wave of innovation and entrepreneurship education, but the evaluation link was obviously weak. This study takes the evaluation of innovation and entrepreneurship education in universities as the research object, and the purpose of this study is to further improve the innovation and entrepreneurship education system in universities [3].

After the Ministry of Education proposed the implementation of innovation and entrepreneurship education in colleges and universities, academic circles related to innovation and entrepreneurship education courses, teachers, teaching and other research have achieved more results, which provides theoretical support for the practice of innovation and entrepreneurship education in universities. However, there are few studies on the evaluation of innovation and entrepreneurship in universities, which makes it difficult for government education administration departments or universities to make timely and effective judgments on the implementation of innovation and entrepreneurship education. This study attempts to construct an evaluation system and model for innovation and entrepreneurship in colleges and universities to provide a complete and practical tool for government education authorities and universities to evaluate the implementation of innovation and entrepreneurship education.

2 Related work

The literature [4] divided entrepreneurship education courses into “gravity courses” and “radiation courses”. The former refers to the common entrepreneurship education courses offered by each major in the same school, and it can also provide education services to students in other schools. The latter refers to entrepreneurial education courses scattered across multiple schools, which are undertaken by multiple educational subjects. The evaluation of the entrepreneurship education curriculum mainly evaluates the learning outcomes of students. The literature [5] took the Hague Vocational Education University in the Netherlands as a case study and confirms that innovation and entrepreneurship education courses have a greater impact on the quality of innovation and entrepreneurship education. The study believed that through the creation of innovation and entrepreneurship courses, “student-centered education” can be incorporated into the innovation and entrepreneurship education system, and various evaluation indicators can be set to evaluate students’ potential for innovation and entrepreneurship and their ability to avoid risks of innovation and entrepreneurship. The research in the literature [6] believed that the entrepreneurial education network plan implemented in the United States integrates innovative entrepreneurship concepts into professional courses, which can cultivate students’ innovative entrepreneurial thinking and problem-solving skills. Based on the students ‘internship results in school or business, students’ creative thinking and cooperation ability can be evaluated.

The research in the literature [7] believed that the teacher factor is the most critical when implementing innovation and entrepreneurship education. The evaluation of the level of innovative and entrepreneurial teachers can be implemented from three dimensions: teacher education, innovative thinking and practical ability. The literature [8] studied the impact of the qualifications of teachers of innovation and entrepreneurship education on the quality of innovation and entrepreneurship education, and believed that the attitudes and self-efficacy of teachers ‘innovation and entrepreneurship education significantly affected students’ motivation for innovation and entrepreneurship and their ability to innovate and start an enterprise. A survey of 315 teachers’ attitudes to innovation and entrepreneurship education and self-efficacy in a technical vocational middle school in Malaysia shows that the vocational middle school teachers have a high sense of self-efficacy in innovation and entrepreneurship, but there is still much room for improvement in terms of self-esteem, personal control cognition, and innovation and entrepreneurial behavior. The literature [9] evaluated the quality of teachers of innovation and entrepreneurship education in agricultural specialty and believed that the innovation and entrepreneurship consciousness and practical ability of agricultural teachers should be used as evaluation indicators of innovation and entrepreneurship education. The quality of teachers ‘innovation and entrepreneurship and the implementation of innovation and entrepreneurship education directly affect the cultivation of students’ awareness of innovation and entrepreneurship and the formation of innovation and entrepreneurship qualities. Therefore, cultivating an excellent teaching team of innovative entrepreneurship education is the key to the implementation of innovative entrepreneurship.

The literature [10] believed that the evaluation of innovation and entrepreneurship education should not be limited to the content of “remembering”. Moreover, this literature believed that we should examine students ‘ability to respond and advanced thinking, and can use group reports, research papers, case studies, business plans or strategy development to evaluate graduates’ innovative entrepreneurship. The literature [11] considered that the evaluation of innovation and entrepreneurship education is a component of education evaluation and evaluated graduate entrepreneurship education through the GIM program. The evaluation indicators of this study include aspects of student behavior, innovation intention, knowledge acquisition and skill return, and the study involves process evaluation and result evaluation. The literature [12] explained the evaluation system of entrepreneurship education in India and believed that Indian business school regards entrepreneurship as a basic course of business education, and covers all aspects of self-employment, joint entrepreneurship and internal entrepreneurship, which has promoted knowledge creation to a certain extent. The literature [13] established an evaluation system for entrepreneurship education practice and believed that the gap between entrepreneurship education in the United States and the United Kingdom was small, and graduate students focused more on entrepreneurship education skills and practice than undergraduates. Reference [14] draws 10 cases based on the database curriculum of 7 universities in Denmark. After case analysis, formative evaluation, learner-centered evaluation, and summative evaluation are considered to be the three main forms of innovation and entrepreneurship education evaluation.

3 C4.5 Algorithm

After Quinlan proposed the ID3 algorithm in 1986, he also proposed an improved version of ID3 in 1993, that is, the C4.5 method. The C4.5 algorithm uses the information gain rate to replace the information gain. The gain ratio (GainRatio) of attribute A is defined as [15]: $GainRatio (A) = Gain (A) Splitlnfo (A)$

In the formula, Gain (A) is the information gain of attribute A, and Splitlnfo (A) is the split information of attribute A. The algorithm uses the information gain rate to select the decision attributes. For each attribute of the data in the training set, the information gain rate is calculated. The attribute with the highest information gain is selected as the test attribute for a given set.

The core idea of the C4.5 algorithm decision tree algorithm is to use the principle of information entropy to select the attribute with the largest information gain rate as the classification attribute, recursively construct the branches of the decision tree, and complete the construction of the decision tree [16].

We set the information entropy of the attributes of n counter examples and p positive examples as: $E (A) = \sum_{i = 1}^{n} \frac{p_{i} + n_{i}}{p + n} I (n_{i}, p_{i})$ (1)

Among them,

$I (n_{i}, p_{i}) = - \frac{n_{i}}{p_{i} + n_{i}} lg \frac{n_{i}}{p_{i} + n_{i}} - \frac{p_{i}}{p_{i} + n_{i}} lg \frac{p_{i}}{p_{i} + n_{i}}$ (2)

$E (A) = \frac{1}{(n + p) {In}_{2}} \sum_{i = 1}^{n} (\begin{matrix} - p_{i} In \frac{p_{i}}{p_{i} + n_{i}} - n_{i} In \\ \frac{n_{i}}{p_{i} + n_{i}} \end{matrix})$ (3)

The C4.5 algorithm improves the ID3 algorithm from the following aspects:

Missing data: When building a decision tree, we can simply ignore the missing data, that is, only the records with attribute values are considered when calculating the gain ratio. To classify a record with a missing attribute value, we can predict the missing attribute value based on other records with known attribute values [17].

Continuous data: The basic idea is to divide the data into regions based on the attribute values of the tuples in the training sample.

Pruning: In the C4.5 algorithm, there are two basic pruning strategies:

Subtree substitution method. It refers to replacing the subtree with leaf nodes. Only when the error rate after the replacement is close to the error rate of the original tree, the leaf node replaces the subtree. Subtree replacement is performed from bottom to top, that is, from the bottom of the tree to the root of the tree [18].

Subtree ascent method. It refers to replacing the subtree with the most commonly used subtree. The subtree rises from its current position to the higher node in the tree. For this replacement, it is also necessary to determine the increase in the error rate.

Rules: The C4.5 algorithm can be classified using either a decision tree or a rule generated from a decision tree. In addition, some techniques have been proposed to simplify complex rules. If all records in the training set are treated equally, the simpler form is used to replace the left part of a rule. When no other rules are available, “other” types of rules can be used to indicate what should be done.

Splitting: The ID3 algorithm favors attributes with more values, which may lead to overfitting. In extreme cases, if an attribute has a unique value for each tuple in the training set, the attribute is considered to be the best, because there is only one tuple for each partition. In order to achieve the best split, the C4.5 algorithm first calculates the gain of each attribute. Then, only those attributes that are higher than the average value of the information gain are tested by applying the gain ratio, and the case where the size of a subset is close to the size of the initial set and the gain ratio is very large is compensated [19].

The C4.5 algorithm has both inheritance and improvement based on the ID3 algorithm, and has the following advantages:

It makes up for the disadvantage of tending to select attributes with more values when selecting attributes with information gain and uses the information gain rate to select attributes.

It continuously adopts pruning in the process of constructing a decision tree.

It can process incomplete data. If an instance cannot be assigned to any branch due to the missing value problem, it can be assigned to each branch according to the weight ratio. The weight ratio is obtained by normalizing the number of training instances contained in each branch with the number of all known training instances contained in the node.

It can take discrete processing on continuous attributes.

Although the C4.5 algorithm has many advantages including the above characteristics, it also has its shortcomings and also has disadvantages when dealing with some problems, mainly including:

The C4.5 algorithm uses a divide-and-conquer strategy, and the choice made in the process of constructing the tree is a locally optimal search method. Although the created decision tree has high accuracy, it does not achieve the overall optimal result [20].

The C4.5 evaluation decision is mainly based on the error rate of the decision tree and does not consider the number of nodes and the average depth of the tree. The average depth of the tree has an important impact on the prediction speed of the decision tree, and the size of the tree is determined by the number of nodes in the tree.

The construction of the decision tree and evaluation are performed simultaneously. Therefore, after the decision tree is constructed, the structure, content, and performance of the tree are difficult to modify.

The C4.5 algorithm tries to group attribute values one by one, which results in low efficiency of grouping.

The C4.5 algorithm is only suitable for data sets stored in memory. The program cannot run when the training set is too large to fit in the memory.

4 C4.5 algorithm improvement

In comparison, the classification rules generated by the C4.5 algorithm are easy to understand, easy to accept, and have high accuracy, but the data set needs to be continuously scanned and sorted during the process of constructing the tree, resulting in low algorithm efficiency. In addition, the C4.5 algorithm needs to calculate the logarithmic function multiple times, and the calculation is difficult, and the calculation time is long. If we look for rules in the calculation process and improve the calculation method, we can greatly save time and simplify operations to increase the speed of creating decision trees. The improvement process is as follows [21 –23]:

We set the information entropy of the attributes of n counter examples and p positive examples as: $E (A) = \sum_{i = 1}^{n} \frac{p_{i} + n_{i}}{p + n} I (n_{i}, p_{i})$ (4)

Among them,

$I (n_{i}, p_{i}) = - \frac{n_{i}}{p_{i} + n_{i}} {lg}_{2} \frac{n_{i}}{p_{i} + n_{i}} - \frac{p_{i}}{p_{i} + n_{i}} {lg}_{2} \frac{p_{i}}{p_{i} + n_{i}}$ (5)

$E (A) = \frac{1}{(n + p) {In}_{2}} \sum_{i = 1}^{n} (\begin{matrix} - p_{i} In \frac{p_{i}}{p_{i} + n_{i}} \\ - n_{i} In \frac{n_{i}}{p_{i} + n_{i}} \end{matrix})$ (6)

Because the calculation result of $\frac{1}{(n + p) {In}_{2}}$ is constant, each factor needs to be calculated repeatedly, which can be ignored. Therefore, E (A) can be improved to: $E (A) = \sum_{i = 1}^{n} (- p_{i} In \frac{p_{i}}{p_{i} + n_{i}} - n_{i} In \frac{n_{i}}{p_{i} + n_{i}})$ (7)

The above formula is used as the basis for selecting the node attribute entropy. According to the principle of mathematically equivalent infinitesimals, if X is very small, then In (1 + X) ≈ X, Available: $In \frac{p_{i}}{p_{i} + n_{i}} = In (1 - \frac{n_{i}}{p_{i} + n_{i}}) \approx - \frac{n_{i}}{p_{i} + n_{i}}$ (8)

$In \frac{n_{i}}{p_{i} + n_{i}} = In (1 - \frac{p_{i}}{p_{i} + n_{i}}) \approx - \frac{p_{i}}{p_{i} + n_{i}}$ (9)

When Equations (8) and (9) are substituted into Equation (7), E (A) becomes: $\begin{matrix} E^{'} (A) = \sum_{i = 1}^{n} (- p_{i} In \frac{p_{i}}{p_{i} + n_{i}} - n_{i} In \frac{n_{i}}{p_{i} + n_{i}}) \\ \approx 2 \sum_{i = 1}^{n} \frac{n_{i} p_{i}}{p_{i} + n_{i}} \end{matrix}$ (10)

$\begin{matrix} S^{'} (S, A) = - \frac{n_{i}}{p_{i} + n_{i}} {lg}_{2} \frac{n_{i}}{p_{i} + n_{i}} - \frac{p_{i}}{p_{i} + n_{i}} \\ {lg}_{2} \frac{p_{i}}{p_{i} + n_{i}} \approx \sum_{i = 1}^{n} \frac{p_{i}}{p_{i} + n_{i}} \end{matrix}$ (11)

$G - R^{'} (S, A) = \frac{I (p, n) - E^{'} (A)}{S^{'} (S, A)}$ (12)

Therefore, we can replace E (A), Split (A), and GainRatio (A) in the C4.5 algorithm with Equations (10), (11), and (12), and select the attribute with the largest “information gain rate” as the node., The improved algorithm eliminates the logarithmic operation, so the algorithm can only perform the addition, subtraction, multiplication, and division operations to calculate the information gain rate, which is fast and efficient.

The basis for algorithm improvement: During the discretization of continuous value attributes, it takes more time to test all the partitions. Therefore, how to choose the best partition point becomes the key to the problem. Fayyad and Irani proved that no matter how many categories and the distribution of the categories of the dataset used for learning, the best dividing point of continuous value attributes is always at the boundary point.

The core idea of C4.5 algorithm is to select the attribute with the largest information gain rate as the classification attribute of the decision tree, and then construct the branches of the decision tree one by one. The algorithm process is: Suppose the training set X of the instance, this instance has m attributes can be divided into n categories, respectively {A₁, A₂, ⋯ , A_m }, {C₁, C₂, ⋯ , C_n }. The training set can be divided into m subsets {X₁, X₂, ⋯ , X_m } by attributes, and it can be divided into n subsets {X₁, X₂, ⋯ , X_n } by categories. Among them, the attribute corresponding to X_i is A_i (i = 1, 2, ⋯ , m) and the category corresponding to X_j is C_j (j = 1, 2, ⋯ , n). We assume that |A_i| represents the number of A_i, |C_j| is the number of instances of C_j, and |X_ij| is the number of instances with attribute A_i and category C_j.

(1) The probability that an instance attribute is A_i is: $P (A_{i}) = \frac{| A_{i} |}{| X |}$ (13)

(2) The probability that an instance belongs to class j is: $P (C_{i}) = \frac{| C_{i} |}{| X |}$ (14)

(3) The probability that the instance of attribute A_i contains category C_j is: $P (C_{j} | A_{i}) = \frac{| X_{ij} |}{| A_{i} |}$ (15)

From Equation (14), we can know that the uncertainty of the decision tree partition C can be obtained as:

$info (X) = - \sum_{i = 1}^{n} P (C_{j}) {log}_{2} P (C_{j}) = - \sum_{| X |}^{| C_{j} |} lg \frac{| C_{j} |}{| X |}$ (16)

From the Equations (13) and (16), we can know that the entropy of the classification information of each leaf node with attribute A_i is: $\underset{A}{info} (X) = \sum_{i = 1}^{m} \frac{| A_{i} |}{| X |} info (A_{i})$ (17)

The information gain rate of attribute A is the classified mutual information amount of attribute A: $Gain (A) = ino (X) - {inofo}_{A} (X)$ (18)

The split information of attribute A is:

$\begin{matrix} Splitinfo (A) & = \sum_{i = 1}^{m} P (A_{i}) {log}_{2} P (A_{i}) \\ = \sum_{i = 1}^{m} lg \frac{\frac{| A_{i} |}{| X |} {log}_{2} | A_{i} |}{| X |} \end{matrix}$ (19)

The gain ratio is used instead of gain. The gain ratio is: $GainRatio (A) = \frac{Gain (A)}{Splitinfo (A)}$ (20)

5 Improved algorithm performance

1. Taylor series

In mathematics, the Taylor series is a continuous addition of infinite terms, that is, a series to represent a function. These added terms are obtained from the derivative of the function at a certain point. It is defined as follows: A Taylor series of infinitely differentiable real or complex function f (x) in the neighborhood of a (a is real or complex) is a power series as follows: $f (x) = \sum_{n = 0}^{\infty} \frac{f^{(n)} (a)}{n!} {(x - a)}^{n}$ (21)

In the formula: n ! represents the factorial of n, and f⁽ⁿ⁾ (a) represents the n-th derivative of function f at point a. If a = 0, this series is also called McLaren series.

By definition, when f (x) is a natural logarithm, that is, when f (x) = In (x), its Taylor series is: $In (1 + x) = \sum_{n = 0}^{\infty} \frac{{(- 1)}^{n + 1}}{n} x^{n}$ (22)

According to the principle of equivalent infinitesimals, when the value of X is very small, the values of $\sum_{n = 2}^{\infty} \frac{{(- 1)}^{n + 1}}{n} x^{n}$ are all approaching 0, that is, $In (1 + x) \approx x$ (23)

2. Simplification of C4.5 algorithm based on Taylor series

Simplify calculations: In the traditional C4.5 information entropy calculation, due to ${log}_{2} P (C_{j}) = \frac{In (C_{j})}{In 2}$ , according to the principle of the above formula, the following formula can be obtained by simplifying: $\begin{matrix} info (X) = - \sum_{i = 1}^{n} P (C_{j}) {log}_{2} P (C_{j}) = \\ - \sum_{i = 1}^{n} P (C_{j}) \frac{InP (C_{j})}{In 2} = \frac{InP (C_{j})}{In 2} \sum_{i = 1}^{n} P (C_{j}) \\ = - \frac{In \frac{| C_{j} |}{| X |}}{In 2} = \frac{1}{| T | In 2} \sum_{i = 1}^{n} \frac{| C_{j} | \times (| T | - | C_{i} |)}{| T |} \end{matrix}$ (24)

In the same way, other formulas related to logarithmic operations are simplified accordingly. Finally, the information gain ratio formula is:

$\begin{matrix} GainRatio (A) = \frac{Gain (A)}{Splitinfo (A)} = \\ - \frac{\sum i = 1 n \frac{| C_{i} | \times (| T | - | C_{i} |)}{| T |} - \sum_{j = 1}^{m} \sum_{i = 1}^{n} \frac{| C_{i} | \times (| A_{j} | - | C_{i} |)}{| A_{j} |}}{\sum_{j = 1}^{m} \frac{| A_{j} | \times (| X | - | A_{j} |)}{| X |}} \end{matrix}$ (25)

Because errors occur during the simplification process, the above formula cannot be used directly to calculate the information gain rate. To compensate for the error, the number of attribute values M of each attribute is added, that is, the information gain rate is multiplied by M: $GainRatio (A) = \frac{Gain (A)}{Splitinfo (A)} \times M$ (26)

The above formula is the final calculation formula after simplification based on the principle of Taylor series and equivalent infinitesimal. Because the calculation formula of the information gain rate does not involve logarithmic operation, the classification efficiency of the algorithm is improved.

1. Introduce balance factor

The C4.5 algorithm uses the information gain rate to replace the information gain based on the ID3 algorithm and reclassifies the attributes. This method can effectively overcome the disadvantage of biasing to multi-valued attributes when selecting attributes due to information gain in the ID3 algorithm. However, in essence, this method only makes the constructed decision tree achieve local optimization in the selection of internal nodes, which can improve the accuracy to a certain extent, but it is difficult to achieve global optimization due to the limitations of the method itself. Moreover, in the C4.5 algorithm, classification information of large classes may also hide classification information of small classes that are not obvious enough. Although the multi-value bias problem is solved by changing the information gain rate, the meaning of the information theory is blurred, and the interpretability is reduced.

Aiming at the shortcomings of the branching strategy and attribute selection of the C4.5 algorithm, this paper introduces a balance factor λ related to split attributes and class attributes to modify the split information of attributes, and then select more meaningful attributes as split nodes to coordinate the information gain rate of each attribute.

The following formula is used to define the balance factor λ: $λ = (\sum_{j = 1}^{n} \sum_{i = 1}^{m} | S_{ij} - E_{ij} |) / (2 * N)$ (27)

It can be known from the above formula that the value of the balance factor λ depends on the classification and class attributes, that is, the two major variables A and C will determine its value. S_ij is the total number of instances of attribute A_i in category C_j. The expected value that attributes and categories are irrelevant is: $E_{ji} = \frac{S_{i *} * S_{* j}}{N}$ (28)

2. Optimization of C4.5 algorithm based on balance factor

The improved C4.5 algorithm is an improvement on the attribute selection criteria. The basic principle of the improved C4.5 algorithm: In the case of optimal attribute partitioning, a balance factor is introduced in the attribute selection metric instead of the traditional C4.5 algorithm, and attribute selection is performed only based on the highest information gain rate, thereby reducing the information entropy of some attributes, adjusting the attribute value information gain rate, and improving the decision tree. Under the equilibrium condition of the attribute A, the modified classification information entropy is: $info * (A) = - \sum (| A_{i} | / | A | + λ) * info (A)$ (29)

The information gain is corrected as: $gain * (A) = | info (A) - info * (A) |$ (30)

The information gain ratio is corrected as: $gainRatio * (A) = Gain * (A) / SplizInfo (A)$ (31)

The pseudo code of the improved C4.5 algorithm is described as follows:

Algorithm analysis: C4.5 improved algorithm: Based on the given training set A, a decision tree is generated. Input: Training set A with category C distinction, Output: a decision tree. The specific implementation steps are as follows

(1) A node N is created;

(2) If instances X are all of the same class C, then N is returned as a leaf node, and N is recorded as class C;

(3) If the attribute set is empty, then N is returned as a leaf node, and N is regarded as the majority class in X;

F (A, X) is called, the attribute A_i is selected from the attribute set A according to the maximum information gain rate, and the attribute A * i with the largest number is counted;

(4) Whether correction is needed is judged and the corresponding operation is performed

if A_i = A_i*

Then

By formula (28), the expected value of the association between the attribute and the category is obtained;

By formula (27), the balance factor is obtained;

By formula (30), the information gain of A after correction is calculated;

By formula (31), the information gain ratio of A after correction is calculated;

Else

The attribute A is selected as the best splitting attribute, and marks leaf node N with attribute A;

(5) A_i value of Foreach property set A

According to the attribute set A, the sample set X is divided into m subsets {X₁, X₂, ⋯ , X_m }.

A subset X_i with attributes A_i is selected, where (i = 1, 2, ⋯ , m)

Whether it is suitable to generate a new subtree is judged.

If X_i is empty

Then a new leaf node is added, labeled as the majority class in X;

Else

F (A, X) is executed on this leaf node.

EndFor

End

The processing flowchart of the algorithm is shown in Fig. 1. The C4.5 algorithm first calculates the information entropy to obtain the gain of each attribute information, then selects the test attribute and selects the attribute with the largest gain value to split and performs recursion until no new node appears.

Fig. 1

Algorithm flowchart.

6 Model construction

Following the principle of analytic hierarchy process, the questionnaire survey sets the importance level of the evaluation indicators of higher vocational innovation and entrepreneurship education to 5 levels such as “1, 2,..., 5”. The larger the value, the greater the importance. After the questionnaire was retrieved, the importance values of each evaluation index judged by 12 experts were counted, and the average value of importance was obtained (three decimal places are retained). If the ratio of the importance of indicator 1 to indicator 2 is a, then the importance of indicator 2 to indicator 1 is the inverse (1 / a) of a. Based on this, a pairwise comparison judgment matrix (or “index weight resource allocation table”) is constructed, and specific results are obtained using the characteristic root method. This study takes the first-level indicator as an example, judges the matrix, and calculates the single order of the first-level indicator.

The average random consistency index is shown in Fig. 2.

Fig. 2

The average random consistency index.

This article chooses a linear weighted method to make a comprehensive evaluation of the index. The specific formula is shown in following. In the formula, Y is the comprehensive evaluation score, X_i is the scoring value of the i-th third-level indicator of a school, and W_i is the comprehensive weight of the i-th third-level indicator.

The evaluation model is constructed as: $Y = \sum_{i = 1}^{28} X_{i} \times W_{i}$ (32)

Table 3 shows the weights and consistency tests of the first-level indicators.

Table 1

Education evaluation index system

First level index	Two level index
Innovation and entrepreneurship education environment c	External support environment C1
	School implementation environment C2
Investment in innovation and entrepreneurship education I	School entrepreneurship C3
	Faculty building I1
	Funding status I2
	Practice platform construction i3
Innovation and entrepreneurship education process P	Curriculum system design P1
	Service guidance support P2
Results of innovation and entrepreneurship education p*	Student participation process P3
	Social impact p * 1
	Education effect p * 2

Table 2

Judgment matrix of primary indicators

Evaluation of Innovation and of results Entrepreneurship Education	Environmental Assessment	Input evaluation	Process evaluation	Evaluation
Environmental Assessment	1	1.076	1.058	1.06
Input evaluation	0.93	1	0.984	0.985
Process evaluation	0.945	1.017	1	1.002
Evaluation of results	0.944	1.015	0.998	1

Table 3

Weights and consistency test results of the first-level indicators

Evaluation of Innovation and Entrepreneurship Education	Environmental Assessment	Input evaluation	Process evaluation	Weights
Environmental Assessment	1	1.076	1.058	0.2531
Input evaluation	0.93	1	0.984	0.2591
Process evaluation	0.945	1.017	1	0.2472
Evaluation of results	0.944	1.015	0.998	0.2406

The three levels of indicators are shown in Table 4.

Table 4

Statistical table of three levels of indicators

Government support policy	C11
Social assistance	C12
Talent training program	C21
Organization and management organization	C22
Establishment of distribution system of entrepreneurial income	C23
Signing of technology transfer	C33
Distribution of teachers in school	I11
Allocation of external entrepreneurship tutors	I12
Allocation of full-time teachers	I13
Proportion of teachers with entrepreneurial experience	I14
Special fund guarantee in school	I21
Personal input of students	I22
Number and scale of practical teaching bases	I31
Number and scale of holding	I32
Opening situation of teachers and students in practice base	I33
Proportion of courses in this major	P11
Proportion of course hours in this major	P12
Proportion of course credits in this major	P13
Penetration of courses in professional courses	P14
Information release of innovation and Entrepreneurship	P21
Construction of innovation and entrepreneurship guidance institutions	P22
Construction of innovation and entrepreneurship education community	P23
Course attendance	P31
Students’ participation in activities	P32
Number of successful Alumni	P*11
Number of enterprises cultivated by the base	P*12
Improvement of students’ Entrepreneurship quality	P*21
Proportion of graduates and entrepreneurs in employed students p * 22	P*21

The secondary and tertiary indicators of the comprehensive weight of the evaluation indicators of innovation and entrepreneurship education in colleges and universities are shown in Figs. 3 and 4, respectively.

Fig. 3

Statistical table of secondary indicator weights.

Fig. 4

Statistical table of three-level indicator weights.

Through comparative analysis of 5 universities, the statistical results were obtained. The evaluation results (first-level indicators) of innovation and entrepreneurship education in five universities are shown in Fig. 5.

Fig. 5

Educational evaluation results of primary indicators.

Fig. 6

Evaluation results of innovation and entrepreneurship education in five universities (secondary indicators).

Fig. 7

Evaluation results of innovation and entrepreneurship education in five universities (third-level indicators).

As far as the secondary indicators are concerned, in the environmental assessment, the external support environment is at the “general” level, the school implementation environment is at the “good” level, and the school entrepreneurial ability is at the “poor” level. Specifically, the school’s social funding for innovation and entrepreneurship education is near “general” level, and the level of innovation and entrepreneurship technology transfer is “poor”, so the distribution system of entrepreneurial income needs to be further improved. In addition, the local government is more supportive of the school’s innovation and entrepreneurship education. The establishment of the school’s internal innovation and entrepreneurship education funding management institutions can basically meet the needs of the school, and the innovation and entrepreneurship talent training program is well developed.

In the input evaluation, the construction of the teaching staff and the construction of the practical platform are close to the “good” level, and the current status of funding input is the “general” level. Specifically, the proportion of teachers with entrepreneurial experience is low, the students’ personal investment is “average”, and the number of external tutors and the number and scale of practical teaching bases are at the “general” level. In addition, some bases are open to teachers and students in the school, and the school’s innovation and entrepreneurship teaching configuration is close to “very good” level. In the process evaluation, the curriculum system design and service guidance support are at the “general” level, and the student participation process is near the “good” level. Specifically, the number of courses, hours, and credits for innovation and entrepreneurship are still low, and the number of innovation and entrepreneurship societies is relatively small, which is close to the “general” level. In addition, the university releases more information on innovation and entrepreneurship, and it is better for students to participate in innovation and entrepreneurship courses at the prescribed time and place. In the results evaluation, the effectiveness and social impact of innovation and entrepreneurship education are close to the “general” level. Specifically, the school’s innovation and entrepreneurship base have a relatively low number of enterprises and graduates’ entrepreneurship. The quality of students’ innovation and entrepreneurship needs to be further improved, and the number of successful graduates’ innovation and entrepreneurship in previous years is also at “average” level.

7 Analysis and discussion

The evaluation process and results show that each sample university attaches great importance to employment education and service work and provides diversified employment guidance and services for students. Through the implementation of multiple employment education activities, students’ employment concepts have been effectively changed, and their job-seeking ability has been improved. In addition, the school promotes entrepreneurship and employment by offering employment services and career planning courses. Through the establishment of entrepreneurship training bases, innovation and entrepreneurship education and training work has been extensively carried out for the society. Some universities have actively cooperated with local governments and related departments to implement free innovation and entrepreneurship training policies, entrepreneurial rewards policies for successful entrepreneurs, and small loan interest-free policies, etc., which have continuously improved the scale and capabilities of innovation and entrepreneurship education in universities. At present, colleges and universities have basically established a complete service system for innovative entrepreneurship education that understands the needs before training, ensures quality during training, and tracks services after training.

Through training, various levels of government’s support policies for university student students’ entrepreneurship are effectively transmitted to each student, and one-on-one guidance is provided for students preparing for entrepreneurship, which effectively stimulates university student students’ entrepreneurial enthusiasm, enhances their ability to innovate and start their own businesses, and promotes their employment and employment. Colleges and universities attach great importance to the cultivation of university students’ innovative consciousness and innovation ability, set up the employment and entrepreneurship education center, and build a “studio+project team” and “research room+project team” innovation and entrepreneurship training mode. In addition, it forms creative experiments and verification under the guidance of a mentor, selects outstanding innovation and entrepreneurship projects from practice to cultivate and incubate and provide external services.

The evaluation process and results show that colleges and universities incorporate students’ innovation and entrepreneurship training into the curriculum system, so that innovation and entrepreneurship education runs through the entire education process. Some case schools have also established classes of excellence and adopted the “3 + X” training model to provide targeted training and entrepreneurial practices to selected students. In addition, a comprehensive innovation and entrepreneurship park has been established to provide free office space, incubation platforms, and market docking conditions for students to start businesses for free. At the same time, we select entrepreneurs, investors, and entrepreneurial successors to form a mentor team to provide students with policy consulting, communication and collaboration skills training, project development, and entrepreneurship guidance. Through the establishment of the university student innovation and entrepreneurship club, a number of professional societies have been established to actively organize university students ‘scientific and technological innovation activities, innovation and entrepreneurship lectures, etc., to improve students’ innovation ability.

The problems are as follows: At present, the vocational students’ desire for innovation and entrepreneurship is relatively strong, but the scale of implementation of innovation and entrepreneurship education cannot meet the needs of students with entrepreneurial desire. At the school level, innovation and entrepreneurship education does not necessarily require students to start their own companies, and anyone who can participate in a company’s entrepreneurship or learn from a startup company can also be called entrepreneurial behavior. Through some applied projects, vocational students can develop innovative thinking and entrepreneurship, such as participating in business activities such as the Internet of Things and e-commerce.

In the process of implementing innovation and entrepreneurship education, some schools are generally facing the dilemma of scattered campuses, limited school size, lack of innovation and entrepreneurship environment, and lack of teaching resources. In addition, innovation and entrepreneurship education has not really become a discipline in China. Although there are individual master’s or doctoral research directions in entrepreneurship, they all rely on pedagogy or management. At present, the degree of entrepreneurship education or the degree of entrepreneurial business management is still conceived, and the teaching materials for innovation and entrepreneurship need to be improved and unified. However, some developed countries, such as the United States, established entrepreneurship disciplines in the 1950 s, began to cultivate masters or doctorates in entrepreneurship research, and formed a relatively mature innovation and entrepreneurship education system.

Whether it is a school to carry out innovation and entrepreneurship education, teachers to start a business, or organize students to participate in innovation and entrepreneurship competitions, they need the necessary financial support. However, in specific practice, students who can truly realize entrepreneurship and get financing to maintain normal operations are still a minority. The government and education administrative departments mainly support innovation and entrepreneurship through policies and have not set up innovation and entrepreneurship guidance funds specifically for university students and have not established a special innovation and entrepreneurship education funding support system. Therefore, it is difficult for university innovation and entrepreneurship education to break through a single classroom teaching to allow students to truly participate in the practice of innovation and entrepreneurship.

8 Conclusion

Based on the research results of innovation and entrepreneurship education evaluation and education quality evaluation at home and abroad, this study builds an evaluation index system of university innovation and entrepreneurship education based on an improved evaluation model of decision tree algorithm. This article elaborates on the basic theoretical knowledge of data mining, explains the classification methods of data mining, and compares several classification algorithms, and summarizes the advantages and disadvantages of each algorithm. The algorithm in this article is constructed using a decision tree. Then, this research compares and analyzes the C4.5 algorithm before and after the improvement and analyzes it through examples. After determining the evaluation indicators for innovation and entrepreneurship education in universities, the researchers use five universities as examples to analyze and define the definitions of the indicators. On this basis, in accordance with the requirements of the relevant documents of the State Council and the Ministry of Education, based on the evaluation indicators of innovation and entrepreneurship education in colleges and universities, specific evaluation standards for each indicator are determined. In addition, the validity of the research model is confirmed through research.

Footnotes

Acknowledgments

The study was supported by “Topics of the 13th Five-Year Plan for National Business Education and Scientific Research in 2019 (GrantNo. SKJYKT-1969): Research on School-level Topics of Jiangsu Agri-animal Husbandry Vocational College (Grant No. NSF201810)”

References

Kim

S.A.

, Ryoo

H.Y.

and Ahn

H.J.

, Student customized creative education model based on open innovation, Journal of Open Innovation Technology Market & Complexity 3(1) (2017), 6–13.

Powell

, Woodfield

L.A.

and Nevill

A.M.

, Increasing physical activity levels in primary school physical education: The SHARP Principles Model, Preventive Medicine Reports 3 (2015), 7–13.

Annand

, Developing a Sustainable Financial Model in Higher Education for Open Educational Resources, International Review of Research in Open & Distance Learning 16(5) (2015), 1–15.

Zare

, Ghodsbin

, Jahanbin

, et al., The Effect of Health Belief Model-Based Education on Knowledge and Prostate Cancer Screening Behaviors: A Randomized Controlled Trial, International Journal of Community Based Nursing & Midwifery 4(1) (2016), 57–68.

Myers

K.R.

and Fredrick

N.B.

, Team Investment and Longitudinal Relationships: An Innovative Global Health Education Model, Academic Medicine Journal of the Association of American Medical Colleges 92(12) (2017).

Kartowagiran

, Evaluation Model for Islamic Education Learning in Junior High School and Its Significance to Students’ Behaviours, Contemporary Sociology 20(3) (2015), 10–11.

Gordon

and Bursuc

, Law and Entrepreneurship Education: A Proposed Model for Curriculum Development, Journal of Legal Studies Education 35(1) (2018), 123–141.

Zeigheimat

, Ebadi

, Rahmati-Najarkolaei

, et al., An investigation into the effect of health belief model-based education on healthcare behaviors of nursing staff in controlling nosocomial infections, Journal of Education & Health Promotion 5(1) (2016), 23–35.

Gråstén

A.W.A.

and Watt

, A Motivational Model of Physical Education and Links to Enjoyment, Knowledge, Performance, Total Physical Activity and Body Mass Index, Journal of Sports Science & Medicine 16(3) (2017), 318–327.

10.

Méndez-Giménez

, Fernández-Río

and Méndez-Alonso

, Sport education model versus traditional model: Effects on motivation and sportsmanship, Internacional De Medicina Y Ciencias De La Actividad Fisica Y Del Deporte 15(59) (2015), 449–466.

11.

Figaredo

D.D.

, La Organización Red en Educación a Distancia (The Network Organizational Model In Distance Education), Social Science Electronic Publishing 55(3) (2016), 45–56.

12.

Priemer

and Hellwig

, Learning About Measurement Uncertainties in Secondary Education: A Model of the Subject Matter, International Journal of Science and Mathematics Education 16(5) (2016), 11–24.

13.

Sri Sumarni

, Dardiri

and Zuchdi

, The Development of Character Education Model Based on Strengthening Social Capital for Students of State Islamic University (UIN) Sunan Kalijaga, Journal of Education & Practice 67(3) (2015), 23–33.

14.

Eisen

and Eaton

D.C.

, A Model for Postdoctoral Education That Promotes Minority and Majority Success in the Biomedical Sciences, Life Sciences Education 16(4) (2017), ar65.

15.

Sönmez

, Association of Cognitive, Affective, Psychomotor and Intuitive Domains in Education, Sönmez Model, Universal Journal of Educational Research 5(3) (2017), 347–356.

16.

Stanley

Z.D.

, Asfour

L.W.

and Weitzman

, Implementation of a peer-mediated health education model in the United Arab Emirates: Addressing risky behaviours among expatriate adolescents, Al Majallah Al Sihhıyah LI Sharq Al Mutawassit 23(7) (2017), 480–491.

17.

Arif

, et al., Optimization of Communication in VANETs Using Fuzzy Logic and Artificial Bee Colony, Journal of Intelligent & Fuzzy Systems 38(5) (2020), 6145–6157.

18.

Shudofsky

and Ballan

M.S.

, Project for Return and Opportunity in Veterans Education (PROVE): An Innovative Clinical Social Work Field Education Model, Clinical Social Work Journal 43(2) (2017), 102–112.

19.

Zhuang

, Intelligent Simulation of Enterprise Re-innovation Support System Based on System Dynamics, Journal of Intelligent & Fuzzy Systems 38(6) (2020), 6883–6893.

20.

Huang

and Huang

, Personalized Overseas Chinese Education Model Based on Map-Reduce Model of Cloud Computing, International Journal of Emerging Technologies in Learning (iJET) 11(4) (2016), 16.

21.

Restrepo

and Meraz

, Development of an Online Data Applications Course Using an Interprofessional Education Model, Nurse Educator 40(4) (2015), E1.

22.

Jiang

and Rao

, Revealing Engineering Management System and Education Model in Developed Countries for Engineering Management Major Education Improvement in China, Frontiers of Engineering Management 1(4) (2015), 406.

23.

Zahrulianingdyah

, Reproductive Health Education Model in Early Childhood through Education Film “Damar Wulan”, Journal of Education & Practice 16 (2015), 77–82.