Hybrid RF–DNN Model for Automated Assessment and Performance Prediction of EFL Learning Outcomes in Higher Education

Assessment of learning outcomes in English as a Foreign Language (EFL) in higher education is a challenging problem because of the heterogeneity of student abilities and limitations in traditional assessment techniques. The existing systems often rely on single-modality information, which can limit the accuracy and interpretability. This paper addresses those limitations with the introduction of a Hybrid RF–DNN framework for multimodal data integration, which combines textual essays, speech features and behavioural academic indicators for accurate performance estimations in a scalable manner. The design of the system combines Random Forest interpretability of structured feature and the representational power of deep neural networks for language and audio. The experiment results show that the hybrid method exceeds the baseline methods significantly with an accuracy of 98.7%, a precision of 98.4%, a recall of 98.1%, an F1-score of 98.2% and an AUC-ROC of 0.992, respectively for LSP50 dataset on average. Such results demonstrate that the model superiorly captures complex nonlinear correlations in multimodal learning data. The proposed method is more reliable and consistent with higher classification accuracy and has practical value for automatic measurement and early academic intervention. The present work lays a strong groundwork for further intelligent EFL learning analytics solutions by outperforming existing studies in terms of the prediction accuracy and multimodal fusion potential.