Machine Learning and Deep Learning Approaches for Cognitive Reasoning Assessment in Personalized Education: A Systematic Literature Review

Abstract

The significance of cognitive reasoning evaluation is on the rise in contemporary education, where not only the students' acquired knowledge but also the process of information analysis, problem solving, and the application of knowledge to new contexts is considered essential for the evaluation. The rapid development of Artificial Intelligence (AI), especially in the area of machine learning (ML) and deep learning (DL), has opened up new opportunities for building intelligent systems that can help automate reasoning assessment and make it more personalized. New methods like Bidirectional Long Short-Term Memory (BiLSTM) networks, transformer language models, multimodal learning, Explainable Artificial Intelligence (XAI), and Large Language Models (LLMs) have made educational assessment more advanced and multifunctional by processing various kinds of learning data and improving model performance and explainability. This paper discusses a systematic review of 48 peer-reviewed articles published between 2016 and 2026 regarding recent developments in AI-based cognitive reasoning evaluation. The selected studies are discussed in terms of six fundamental research topics, which include traditional machine learning algorithms, sequential deep learning and knowledge tracing, transformer networks, multimodal learning frameworks, explainable AI, and cognitive diagnosis models. A structured review approach is applied in order to analyze and contrast different learning models, methods of explainability, educational data sets, and performance metrics, as well as detect trends in research and existing challenges. The conducted review reveals the evident shift from traditional prediction models to explainable and multimodal, as well as LLM-powered cognitive diagnostic systems. Despite the fact that recent methods demonstrated significant advances in predicting students' performance and offering personalized learning opportunities, there are still many challenges, such as a lack of benchmark data sets, annotation costs, domain generalization, fairness issues, transparency problems, and implementation of explainable AI in real-world settings. Based on these results, the paper proposes possible future research directions for creating reliable, interpretable, and learner-centered cognitive reasoning assessment frameworks.