A Comparative Study of AI Models and AI-Based Approaches for Evaluating Subjective Answers in Exams

Abstract

Education has long relied on manual evaluation methods, but as assessments scale, traditional grading faces challenges of inconsistency, bias, and inefficiency. AI-driven approaches have emerged as promising alternatives, leveraging NLP models, deep learning architectures, and hybrid techniques to enhance grading accuracy and scalability. This paper presents a comparative study of various AI-based models, including BERT-based frameworks, transformers, and recurrent neural networks, analyzing their strengths, limitations, and applications in automated subjective answer evaluation. Key challenges such as contextual understanding, rubric alignment, and fairness in AI-based grading are discussed. This study aims to provide insights into the evolving landscape of automated grading and its potential to transform educational assessments.