This study aims to identify the key architectural components of an AI-driven personalized learning system grounded in mathematical proficiency, to explicate the workflow guiding its design and development, to describe the construction of personalized learning paths that support learner progression and to examine the system's diagnostic efficiency in accurately identifying learning gaps across the mathematical procedures (MAP) and structure of the observed learning outcomes (SLO) dimensions.
The study adopts a design research methodology integrating psychometric modeling through the multidimensional random coefficients multinomial logit model, a decision-tree machine learning algorithm (depth = 3), and adaptive system design. Mathematical proficiency was assessed using the Wright map analysis within a multidimensional Rasch framework to define criterion-based proficiency zones across the MAP and SLO dimensions. Item-level response data from Thai Grade 7 students were used as independent inputs to the decision-tree model to diagnose proficiency levels (Levels 1–5) in each dimension separately, informing the development of expert-validated adaptive instructional designs aligned with cognitive diagnostic assessment principles.
The study develops a scalable intelligent personalized diagnostic and tutoring system (IPDTS) composed of six core modules, including diagnostic assessment, AI-driven personalized learning paths, adaptive domain instruction, personalized feedback, learning profile monitoring and interactive engagement tools. Results from the system development and evaluation phases indicate that the decision-tree–based personalized learning path model demonstrates moderate to strong diagnostic efficiency across MAP and SLO cognitive dimensions, particularly at clearly differentiated proficiency levels. The system effectively identifies learners’ strengths and learning gaps, supports targeted instructional routing and provides interpretable diagnostic feedback. However, reduced diagnostic sensitivity was observed at intermediate proficiency levels, highlighting the need for richer data representation and further refinement during large-scale implementation.
This study proposes a novel AI-driven educational framework that integrates psychometric diagnostics, interpretable AI decision models and adaptive instructional strategies within a unified design research framework. Unlike conventional AI-based learning systems that rely primarily on performance data, the proposed approach emphasizes diagnostically informed personalization aligned with learning progression constructs, supporting scalability and equity in mathematical learning.
