Knowledge Tracing and Student Modeling in Intelligent Tutoring Systems: A Systematic Literature Review of Computational Methods, Learner State Estimation, and Predictive Performance

Computer-based intelligent Tutoring Systems (ITS) use Student Modelling and Knowledge Tracing (KT) at the core of adaptive learning environments to model and predict each learner's mastery. While conversational AI improves pedagogical feedback, the core adaptability of ITS ultimately depends on strong Educational Data Mining (EDM) and Learning Analytics (LA) algorithms. Existing literature reviews often conflate frontend instructional AI with backend analytics, leaving a critical gap in synthesizing the computational approaches and predictive systems that drive these intelligent systems. To address this gap, a Systematic Literature Review (SLR) was conducted following PRISMA guidelines, systematically analysing 15 high-quality studies focused only on the computational dimensions of ITS. The results show a clear model shift away from simple traditional probabilistic models towards the latest versions of deep-learning based classic DKT architectures, including context-aware gating and Transformer networks. Learner state representations are also increasingly complex and add behavioural interaction histories rather than simple testing observations of correct or incorrect outcomes. While the validity of these data-driven models continues to be high (particularly in well-structured domains like mathematics), their real-world operational challenges are much harder to resolve, with major technical issues of data sparsity, algorithmic bias, sustainability of pedagogic fit and the diffuse "black-box" of deep neural networks common across the literature. These issues mean that future researches, defined by the authors as the ongoing interaction between the computational and human, should seek to develop explainable AI (XAI) and fairness-aware algorithms that support scalable, trusted human-AI teaming.