Abstract
Teaching quality evaluation is an important work in teaching management. In order to improve the accuracy of teaching quality evaluation, according to the evaluation data at a certain university, an evaluation model based on BP neural network optimized by gravitational search algorithm (GSABP) is proposed. For the GSA algorithm is easy to fall into the local optimal, the ergodicity of chaotic sequence is used to generate the initial population of GSA, and then the chaotic gravitational search algorithm (CGSA) is presented. The experimental results show that, compared with BP neural network and GSABP algorithm, the model using CGSABP has high credibility and strong generalization ability, which provides a feasible method for the accurate evaluation of teaching quality.
Introduction
Evaluating university teaching quality is a critical yet challenging administrative task. It involves a "multi-variable and complex nonlinear" problem where traditional linear models often fall short. Standard methods typically rely on weighted averages of student surveys, which may not accurately reflect the nuanced interactions between teaching attitude, content, and student outcomes.
Artificial Neural Networks (ANNs), specifically Back-Propagation (BP) networks, offer a powerful alternative due to their ability to model complex nonlinear relationships. However, BP networks are notoriously sensitive to initial weight settings. If initialized poorly, they often converge on "local optima"—suboptimal solutions that hinder the model's overall accuracy. To address this, researchers have developed the CGSABP model to automate and optimize these settings.
Methodology: The CGSABP Model
The proposed solution, CGSABP, integrates three distinct computational techniques to create a robust evaluation framework:
- BP Neural Network: This serves as the primary classification engine. It processes 13 distinct teaching indexes to predict a quality grade.
- Gravitational Search Algorithm (GSA): Inspired by the law of universal gravitation, GSA treats potential solutions as masses that attract one another. This allows the algorithm to perform a global search for the optimal weights and thresholds of the BP network.
- Chaotic Sequences: To overcome GSA's tendency to get stuck in local optima, the authors introduced Chaos Theory. By using a chaotic sine map to initialize the population, the algorithm ensures "ergodicity"—a more thorough exploration of the search space compared to standard random initialization.
Experimental Setup and Data
The study utilized real-world data from a university evaluation system. The dataset comprised 30 groups of evaluation data, split into 18 training samples and 12 test samples. Each sample included 13 input indexes (X1-X13) covering teaching attitude, content, methods, and effects. The output (Y) was categorized into a four-grade scale: 1 (Excellent), 2 (Good), 3 (Qualified), and 4 (Failed).
Results and Discussion
The performance of the CGSABP model was compared against a standard BP network and a basic GSABP model. The results demonstrated a clear hierarchy in accuracy:
- Standard BP Network: 75% accuracy.
- GSABP (Standard GSA + BP): 83.33% accuracy.
- CGSABP (Chaotic GSA + BP): 91.67% accuracy.
As shown in the comparison plot, the CGSABP model's predictions align most closely with the actual categories. Out of the 12 test samples, the CGSABP model only misclassified one instance, whereas the standard BP and GSABP models showed significantly more variance and lower reliability.
The numerical data in the evaluation results table further highlights the precision of the CGSABP approach. The predicted values for CGSABP are consistently closer to the integer-based actual values, indicating superior generalization ability and credibility compared to the other two methods.
Conclusion
The integration of chaotic sequences and the gravitational search algorithm significantly enhances the performance of BP neural networks in teaching quality evaluation. By avoiding local optima and ensuring a more comprehensive search of the parameter space, the CGSABP model provides a highly accurate and feasible method for university administrators. Future research could expand the dataset size to further validate the model's universal applicability across different educational contexts.
Summary updated