Deep learning and fuzzy algorithm-based optimization of online English teaching effect

Abstract

Online English teaching systems face challenges in adapting to diverse learner behaviors and uncertainty in educational data, limiting the effectiveness of personalized instruction in digital environments. This research proposes a hybrid deep learning and fuzzy optimization framework, ModRNNet–IT2FS–AdapCS, to enhance the prediction and evaluation of online English teaching effectiveness through sequential behavior modeling and uncertainty-aware decision-making. The model is evaluated on the Online English Teaching Effectiveness Dataset containing 7,500 student records. Data preprocessing includes outlier removal using the Interquartile Range (IQR) method, feature scaling via Min–Max normalization, and dimensionality reduction using Principal Component Analysis (PCA). ModRNNet captures temporal dependencies in student learning behavior, while Interval Type-2 Fuzzy Systems (IT2FS) handle uncertainty in teaching evaluation. Adaptive Cuckoo Search (AdapCS) optimizes network weights and fuzzy membership parameters to improve convergence and predictive stability. The research was implemented using Python 3.11. The proposed model achieves 96% recall for the High effectiveness class, 95% accuracy, 94% F1-score, and 92% precision. It outperforms the Fuzzy Neural Network–Bayesian Optimized Quality Evaluation (FNN-BOQE) model, improving accuracy from 86% to 95%. Statistical significance was assessed using a paired t-test, and the observed performance improvements were found to be statistically significant (p < 0.05), confirming the reliability and consistency of the proposed framework. The ModRNNet–IT2FS–AdapCS framework provides a robust and adaptive solution for intelligent evaluation of online English teaching effectiveness, improving both prediction accuracy and interpretability in uncertain learning environments.