StanBKT: Rethinking Parameter Estimation in Bayesian Knowledge Tracing

StanBKT introduces an open-source Python package for Bayesian Knowledge Tracing (BKT) that moves beyond traditional expectation-maximization (EM) point estimates to full Bayesian inference via Stan. The package supports four estimation methods (Hamiltonian Monte Carlo, variational inference, Pathfinder, and optimization), three model variants (standard, grouped, and hierarchical BKT), and provides posterior predictive inference with visualization utilities. Evaluated on the ASSISTments 2020 dataset, all methods achieve comparable predictive performance while trading off speed vs. posterior fidelity. A key demonstration shows how posterior inference enables principled comparison of learning, forgetting, guessing, and slipping parameters across experimental conditions in an educational intervention involving perceptual cue manipulations.

For Intelligent Tutoring systems, this represents a significant methodological advance. Traditional BKT implementations produce point estimates that obscure uncertainty and prevent reliable condition comparisons — critical for A/B testing adaptive interventions. The hierarchical modeling support is particularly valuable for Adaptive Learning where student-level variation must be modeled. Compared to Neural Symbolic Knowledge Tracing approaches, StanBKT preserves the interpretable hidden Markov structure of BKT while adding uncertainty quantification. For Knowledge Tracing IRT comparisons, the Bayesian framework provides natural extensions unavailable in IRT. The work complements Mbp KT Meta Behavioral Knowledge Tracing by offering a complementary probabilistic modeling framework. As an Open Source tool, StanBKT lowers the barrier for Learning Analytics researchers to adopt Bayesian methods in educational data mining.

Connected Concepts

Intelligent Tutoring

Adaptive Learning

Open Source

Learning Analytics

Connected Articles

Neural Symbolic Knowledge Tracing

Knowledge Tracing IRT

Mbp KT Meta Behavioral Knowledge Tracing

Citation

Siddhartha Pradhan, Yanping Pei, Morgan Lee, Puyuan Zhang, Erin Ottmar, Adam C. Sales (2026). StanBKT: Rethinking Parameter Estimation in Bayesian Knowledge Tracing. arXiv:2605.23048. arXiv preprint (cs.HC, cs.CY, stat.AP, stat.ME).