AI-Driven Cognitive Grammar Feedback and Writing Skill Trajectory Modeling System for ESL Student Assessment

This study addresses persistent challenges in ESL writing assessment, where traditional methods lack scalability, timely feedback, and deep cognitive grammar insights. Existing studies improve accuracy but still struggle with limited interpretability, inconsistent scoring, and weak long-term learner modelling. To overcome these limitations, the objective is to develop an AI-driven framework combining DeBERTa for cognitive grammar error detection, T5 for personalized feedback generation, and BiLSTM for writing skill trajectory prediction. Implemented using Python and PyTorch, the proposed system is trained on the IELTS-based writing dataset and delivers a 14% improvement in error-detection accuracy, a 12 % increase in feedback quality, and a 10 % rise in trajectory prediction accuracy compared with baseline studies. Results demonstrate superior linguistic precision, reliable score prediction, and enhanced learner progress modelling. Overall, the system offers a scalable, interpretable, and high-impact solution for modern ESL writing assessment.