Design of a Cognitive Digital Twin for Students in STEM Education: Modeling Learning Uncertainty in Mathematics and Physics Integrating Fuzzy Logic and Explainable AI (XAI)

Design of a Cognitive Digital Twin for Students in STEM Education: Modeling Learning Uncertainty in Mathematics and Physics Integrating Fuzzy Logic and Explainable AI (XAI)

Amirhossein Taghizadeh Gharajeh Ghiaei1 Maryam Irandoust2 Fatemeh Zahra Gharosi3 Narges Khadem Niaei4

1) M.Sc. Student in Applied Mathematics (Numerical Analysis specialization), University of Tabriz, Iran –
2) M.Sc. in Solid-State Physics, Urmia University, Iran –
3) Undergraduate Student of Mathematics Education, Farhangian University (Nasibeh Campus), Tehran, Iran –
4) Undergraduate Student of Mathematics Education, Farhangian University (Nasibeh Campus), Tehran, Iran –

IOI Code : AAHD-ABHBHZ
Publication : National Conference on Information Technology, Nanotechnology, Artificial Intelligence and Technological Futures Studies(itnaf.ir)
Abstract :
    Transitioning from outcome-based evaluation to process-oriented and cognitive-based approaches is a fundamental necessity for enhancing educational quality in STEM disciplines. In subjects such as physics and mathematics, learning is not a linear trajectory but an intertwined network of abstract concepts and computational skills, where partial comprehension in one domain precipitates performance degradation in another. Aiming to demystify this cognitive black box and provide an analytical decision-support framework, this study proposes a novel architecture predicated on the concept of a Cognitive Digital Twin (CDT). Within this architecture, the student s perceptual state is continuously simulated as a virtual counterpart synchronized with reality. Given the non-deterministic and ambiguous nature of human learning, the modeling of cognitive variables is executed via Fuzzy Inference Systems (FIS), enabling varying degrees of mastery in foundational concepts (e.g., algebra, analytical geometry, and calculus) to be quantified as continuous membership functions. To circumvent the opacity inherent in deep learning models and foster teacher trust in algorithmic outputs, this research leverages Explainable Artificial Intelligence (XAI) techniques, specifically Shapley Additive exPlanations (SHAP). This mechanism allows the system to compute and interpret the precise marginal contribution of specific mathematical misconceptions to performance deficits in physics (e.g., kinematics and dynamics). For model validation, a synthetic dataset comprising 500 performance records was generated and analyzed using Monte Carlo statistical distributions. Empirical results demonstrate that the proposed hybrid neuro-fuzzy architecture identifies latent learning bottlenecks with 95% accuracy. This achievement not only provides a robust instrument for micro-level pedagogical interventions and personalized learning but also establishes a data-driven foundation for macro-level educational policymaking and curriculum revision.

Keywords : cognitive digital twin stem education fuzzy logic explainable ai (xai) monte carlo simulation cognitive modeling