The teaching of English at colleges encounters difficulties because student achievement ranges widely and students learn through various factors, while teaching outcomes cannot be assessed without bias. The conventional analysis methods, which include descriptive statistics and linear regression, and experience-based teacher assessment, fail to reveal hidden patterns within extensive learning datasets. The research introduces an artificial intelligence framework that evaluates college English performance through its two main components: a multilayer perceptron (MLP) neural network and a random forest algorithm, which processed performance data from 583 undergraduate students. The academic year data set contains anonymous student learning behavior and assessment records, which were collected through a university academic affairs management system and an online learning platform. The analysis used Pearson correlation analysis to select twelve important feature variables. K-means clustering was applied as an unsupervised learning method to divide students into four learning categories: excellent, good, average, and in need of improvement. An MLP regression model with 12 input nodes, two hidden layers containing 64 and 32 neurons, and one output node was constructed to predict continuous English scores using the ReLU activation function and Adam optimizer. A random forest algorithm was employed to quantify the influence of different learning features on performance. Experimental results indicate that the proposed framework achieves a tolerance-based prediction accuracy of 92.3% within ± 5 score points, with a root mean square error (RMSE) of 4.76. Learning time, online quiz performance, and homework completion rate were identified as the most influential factors, demonstrating the framework’s effectiveness in supporting personalized instruction and data-driven teaching decisions.
Jiang et al. (Fri,) studied this question.