Adaptive Testing with Machine Learning: Customizing English Proficiency Assessments Based on Learner Performance

The implementation of adaptive testing for English proficiency assessments using deep reinforcement learning (DRL) to customize evaluations based on learner performance. The dataset utilized is Classification of English Learner Proficiency, consists of essays ranging from 20 to 300 words, categorized into six Common European Framework of Reference (CEFR) levels: A1, A2, B1, B2, C1, and C2. The research emphasizes a thorough data preprocessing phase, which includes text cleaning and normalization to enhance data quality, followed by feature extraction and tokenization to prepare the textual data for analysis. The core of the study presents a proposed adaptive testing method that leverages a DRL agent to optimize question selection, adjusting the difficulty based on real-time learner performance metrics such as accuracy and response time. This adaptive approach not only personalizes the assessment experience but also maintains learner engagement by presenting appropriately challenging content. The evaluation of the proposed system involves comparative studies against traditional testing methods to assess improvements in learner outcomes and satisfaction. Initial findings indicate that the DRL-driven adaptive testing framework significantly enhances the accuracy of proficiency level predictions while reducing assessment time. Key evaluation metrics demonstrate strong performance, with a precision of 0.92, recall of 0.95, an F1-score of 0.90, and overall accuracy reaching 96%. This research highlights the potential of integrating advanced machine learning techniques in language education, paving the way for more responsive and individualized assessment methodologies.