Evaluation and analysis of teaching quality of university teachers using machine learning algorithms

1 Introduction

With the development of society, the scale of university education has been expanding, and the teaching quality of universities has been widely concerned as an important criterion for judging the excellence of universities. Evaluating teachers’ teaching quality can help people grasp the teaching effect of universities and help teachers, students, and school administrators better understand the shortcomings in the teaching process, thus promoting the efficiency of teachers’ teaching and school management. Aiming the evaluation of classroom teaching quality, Cheng and Yan [1] combined a generalized neural network (GRNN) with the fruit fly optimization algorithm to enhance the learning ability of GRNN. They found through experiments that the improved algorithm had smaller errors. Dong et al. [2] used a diversified evaluation method and applied mathematical fuzzy hierarchical analysis to evaluate teachers’ teaching quality scientifically and objectively. Jian [3] designed a set of evaluation indexes suitable for multimedia teaching quality, tested the evaluation results with the genetic algorithm (GA), and verified the method’s effectiveness through experiments. Bao and Yu [4] proposed a teaching quality evaluation method based on mobile edge computing for mixed online and offline sports teaching and found through experiments that the method had a small error and high efficiency. Evaluating and analyzing the teaching quality of university teachers with a reasonable and effective method can provide reliable data support and a theoretical basis for teaching and management work, thus promoting the steady development of university talent training [5]. With the development of technology, machine learning methods have appeared, and decision trees and association rules have been well applied in teaching quality evaluation, proving the reliability of machine learning methods. This work studied the usability of the support vector machine (SVM) method, aiming to further understand the value of the SVM method for evaluating teaching quality. To improve the performance of the SVM algorithm, its parameters were optimized by the GA to obtain a GA-SVM model. Moreover, an experimental analysis was performed to verify the effectiveness of the GA-SVM model. This study combined two machine learning algorithms and utilized the advantages of the GA to optimize the SVM algorithm, contributing to the further application of the SVM algorithm in the field of education. The results of this study verified the reliability of the optimized SVM method for evaluating teaching quality. The method can be applied in the teaching quality evaluation of more courses.

2 Evaluation index of teaching quality of university teachers

Evaluating teachers’ teaching quality can help people understand the quality of teachers and their work and find out the shortcomings in time to improve the quality of teaching teams and the teaching quality of universities [6]. Therefore, teaching quality evaluation and analysis must be objective, fair, and scientific, while many methods used at present have strong subjectivity and arbitrariness. The evaluation of the teaching quality of teachers in universities should use scientific, reasonable, valid, and reliable indicators; therefore, this study selected evaluation indicators according to the following principles:

1. scientific: the chosen indicators should be consistent with the objective laws of teaching and learning;

2. complete: the chosen indicators should be as comprehensive as possible;

3. measurable: the chosen metrics should be measurable, explicit, and able to be grasped intuitively;

4. orientated: the chosen indicators should be in line with modern teaching concepts and conducive to the comprehensive development cultivation of talents;

5. valid: the chosen indicators should effectively evaluate the teachers’ lesson preparation and knowledge reserve.

Based on the above principles and the teaching content of university teachers, this study selected indicators from four aspects, as shown in Table 1.

Table 1 includes 16 indicators, and the value of every indicator is 100 points. All the 16 indicators are applicable to the teaching quality evaluation of teachers who teach different courses, which can be applied school-wide. The evaluation indicators are clear at a glance and easy for students to score.

3 SVM-based evaluation algorithm

The SVM algorithm is a kind of machine learning algorithm [7] that can solve small-sample, nonlinear problems [8] and has been commonly used in pattern recognition [9], computer vision [10], and image classification [11]. Suppose there is a sample set S as follows:

The classification hyperplane is written as:

where w is the weight and b is the bias. The problem of finding the optimal hyperplane is expressed as follows:

However, in practical problems, the data are mostly linearly inseparable [12]; thus, it is necessary to introduce penalty parameter C and relaxation factor ξ , then,

Then, the final optimization problem can be written as follows:

where a is the Lagrangian multiplier, a = ( a i , ⋯ , a l ) , and K ( x i , x j ) is the kernel function, the radial basis function (RBF) in this study is as follows:

where σ is the parameter of the kernel function.

The final classification function of the SVM algorithm can be written as follows:

where sgn represents the sign function and b ' represents the classification threshold value.

The performance of the SVM algorithm is mainly affected by two parameters [13]. In order to further improve the performance of the SVM algorithm, it was improved by using GA in this study. The GA is a stochastic search algorithm [14], which searches for the optimal solution by simulating the natural evolutionary process [15], and has good applications in data prediction [16], industrial control [17], parameter optimization [18], etc. The specific steps of optimizing parameters with the GA are as follows.

1. Initial population: an initial population was established. The binary encoding method was used. Every individual contained two random numbers, i.e., parameters C and σ of the SVM algorithm.

2. Fitness function: the fitness function of every individual, i.e., the minimum mean squared error of the SVM algorithm, was calculated.

3. Selection: the operation was selected using the roulette algorithm [19], and the best individual entered the next generation.

4. Crossover: individuals were randomly paired, and some of the chromosomes were exchanged based on a probability.

5. Mutation: genetic values at some positions were changed based on probability.

6. The computation was repeated for the fitness function until the termination condition was satisfied. Finally, the optimal parameters C and σ of the SVM algorithm were obtained.

The parameters optimized by the GA were input into the SVM model to obtain the GA-SVM model. The GA-SVM model can be applied to evaluate the teachers’ teaching quality. It needs to be trained using training data and then tested to obtain the evaluation results.

4 Experimental analysis

5 Experimental results

The SVM and GA-SVM algorithms were compared with the Naive Bayes (NB) algorithm [21], the K-neural network (KNN) algorithm [22], and the decision tree (DT) algorithm [23], and the results are shown in Table 4.

It was seen from Table 4 that when classifying different datasets, the accuracy of the NB algorithm was below 70%, the lowest, and the accuracy of the KNN and DT algorithms was between 70 and 80%, slightly higher than the NB algorithm; the SVM algorithm had an accuracy above 80%, and its accuracy for the Iris dataset was 88.64%, which was 20% higher than the NB algorithm and 10% higher than the DT algorithm. These results demonstrated that the SVM algorithm had the best classification performance among these machine learning algorithms. Finally, the comparison between the SVM and GA-SVM algorithms showed that when classifying the Iris dataset, the accuracy of the GA-SVM was 97.33, 8.69% higher than the SVM algorithm, verifying that the improved algorithm was effective.

The comparison of the testing time is shown in Table 5.

It is seen from Table 5 that these algorithms all suggested similar features, i.e., the larger the sample size in the dataset is, the longer the testing time is. These algorithms spent the longest time in classifying the Diabetes algorithm and the shortest time in classifying the Iris dataset. The comparison of these algorithms showed that the testing time of the NB, KNN, and DT algorithms was longer, indicating that the SVM was more efficient in computation. The comparison between the SVM and GA-SVM algorithms demonstrated that when classifying the Diabetes dataset, the testing time of the GA-SVM algorithm was 0.78 s, which was 15.22% shorter than that of the SVM algorithm (0.9 s), indicating that the improvement of the SVM algorithm effectively improved the computational efficiency of the SVM algorithm.

First, the time required by the two algorithms for teaching quality evaluation was compared on the actual datasets, as shown in Figure 1.

Figure 1

Comparison of the running time between algorithms.

It was seen from Figure 1 that the running time of the SVM algorithm was slightly longer. For the training of the algorithm, the running time of the SVM algorithm was 27.84 ms, while the time of the GA-SVM algorithm was 23.21 ms, which was 16.63% shorter than the SVM algorithm. For the testing of the algorithm, the running time of the SVM algorithm was 9.88 ms, and the running time of the GA-SVM algorithm was 7.25 ms, which was 26.62% shorter than the SVM algorithm. In the experiments, the difference in the running time between the two algorithms was less significant due to the small amount of data used, but the above results verified that the GA-SVM algorithm was more efficient.

The evaluation results of the two algorithms were compared, and some of the results are shown in Table 6.

It was seen from Table 6 that among the seven samples, the error between the evaluation value of the SVM algorithm and the actual value was large, and three samples were wrongly evaluated. For example, the sample with an “excellent” result was evaluated as “good,” and the sample with a “good” result was evaluated as “poor.” In contrast, the evaluation value of the GA-SVM algorithm was closer to the real value, i.e., more accurate, and only one sample was wrongly evaluated. The results verified that the GA-SVM algorithm was reliable in evaluating the teachers’ teaching quality.

Finally, the evaluation accuracy of the two methods was compared, as shown in Figure 2.

Figure 2

Comparison of the evaluation accuracy between the two methods.

It was seen from Figure 2 that in the ten times of cross-validation, there was a large gap in the accuracy between the two methods, and the accuracy of the SVM algorithm ranged from 80 to 90%, with a maximum of 89.91% and a minimum of 82.78%, the stability of the evaluation results obtained by the SVM algorithm was also not very high, i.e., there were fluctuations, while the accuracy of the GA-SVM algorithm always stayed above 95%, with a maximum of 98.93% and a minimum of 97.89%. The average accuracy of the GA-SVM algorithm was 11.64% higher than that of the SVM algorithm (98.36 vs 86.72%), verifying the effectiveness of the GA-SVM algorithm in evaluating the teachers’ teaching quality. The above results demonstrated that the GA-SVM algorithm could accurately evaluate the teaching quality and be further promoted and applied in practice.

6 Conclusion

This work studied the evaluation and analysis of teaching quality in universities with the SVM algorithm. The SVM algorithm was improved by optimizing the parameters of the SVM algorithm after selecting evaluation indicators. Then, the GA-SVM algorithm was applied to the dataset of the actual evaluation results of teaching quality. The experimental results showed that the GA-SVM algorithm had a shorter running time and obtained results closer to the actual situation. The accuracy rate of the GA-SVM algorithm was 11.64% higher than the SVM algorithm, which verified the method’s reliability. This work provides some theoretical bases for better evaluation of the teaching quality of university teachers.