Research On Evaluation Of Teaching Quality In Colleges And Universities Based On Adaptive BP And DDAE-SVR Neural Network Models

Evaluation of teaching quality in Colleges and universities is an important segment in the process of teaching management.It is affected by many factors,and its evaluation index and teaching quality is a complicated and abstract nonlinear problem.However,there are some problems for existing evaluation methods,for instance,the subjectivity and randomness are too strong,the index weight is difficult to determine,the over-fitting,the convergence speed is slow and weak computing power,etc.Besides,the evaluation index system mostly focuses on teaching attitude,teaching content,teaching methods,etc.It seldom considers the preparation of pro-teaching teachers and the teaching situation of teaching process,which lead to the lack of comprehensive evaluation.Therefore,how to construct a model to evaluate university teaching quality objectively,truthfully,comprehensively and accurately not only helps to improve the quality of teaching,promotes the constant improvement of teaching objectives and enhances the scientific decision-making of education,but also helps promote the standardization and intelligence of university teaching management.In order to solve the complex non-linear problem of evaluation of teaching quality in colleges and universities,this paper deeply studies the neural network model and college teaching quality evaluation theory.Given the inadequacy of existing research,this paper proposes to build an adaptive BP neural network model and DDAE-SVR deep neural network model to evaluate the quality of teaching in Colleges and universities.DDAE-SVR is a deep neural network that deep denoising auto-encoder(DDAE)for unsupervised training and support vector regression(SVR)is used to conduct supervised prediction.The main research contributions of this article are as follows:(1)Put forward an adaptive BP neural network model.The model introduces adaptive learning rate and momentum term to improve the gradient descent method of the BP neural network to enhance the convergence speed of the model and optimizes the network structure to ensure the stability of the model.Then,the new evaluation indexes are added to the traditional evaluation indexes to ensure that the model comprehensively evaluates teaching activities.The normalized evaluation sample data sets as the input feature vector of the model to improve the computational efficiency.Input small-scale,low-dimensional evaluation data sets into the model for training and validation,and MSE,prediction accuracy,and training time are used as model's performance indexes to compare with standard BP neural network,decision tree model,and support vector machine.The results show that the model well overcomes the problems of excessive subjectivity,difficulty in determining weights,over-fitting,slow convergence,and local minimums,although it is weaker in training time than the support vector machine and decision tree model,the other indexes of the model are optimal,which validates the effectiveness of the model in teaching quality evaluation in colleges and universities.(2)Put forward a DDAE-SVR deep neural network model.In the face of high-dimensional large-scale evaluation data sets,since the adaptive BP neural network model has only one hidden layer,its computational power,prediction ability,and modeling expression capability are limited,therefore,this paper proposes a DDAE-SVR deep neural network model.The model has multiple hidden layers,performs multiple feature conversions during unsupervised training,and minimizes the error between the reconstructed output data and the original input data to obtain the essential characteristics of the data.The model's output layer uses support vector regression as a predictor to implement supervised evaluation prediction.Enter the small-scale evaluation sample data set and large-scale data set to experiment with this model,training time,the mean absolute percentage error(MAPE)and the mean square error(MSE)are used as model's performance indexes to compare with other models.The experimental results show,when dealing with small-scale data sets,although the DDAE-SVR model is higher in training time than the Adaptive-BPNN model,it is superior to the Adaptive-BPNN model in other performance indicators,which validates the effectiveness of the DDAE-SVR model.On the other hand,the difference between the error performance index of the Adaptive-BPNN model and the DDAE-SVR model is small,and its training time is much better than the DDAE-SVR model,which validates Adaptive-BPNN model is more suitable for handling small-scale evaluation data sets.For processing large-scale data sets,the DDAE-SVR model's train time is still higher than the Adaptive-BPNN model,but it is far superior to the Adaptive-BPNN model in terms of error performance.It is verified that the DDAE-SVR model not only has powerful computing ability and modeling ability,but also proves to some extent that it has good prediction accuracy and convergence.