SISTEM ASESMEN ADAPTIF BERBASIS AI MENGGUNAKAN Q-LEARNING UNTUK PEMBELAJARAN PEMROGRAMAN BERBASIS OUTCOME-BASED EDUCATION

Abstract

This study develops an AI-Based Adaptive Assessment System Using Q-Learning for programming learning within an Outcome-Based Education (OBE) framework. Conventional assessments often expose all learners to the same sequence of items, which may provide limited diagnostic information about individual competency. The proposed system integrates an OBE knowledge base, Large Language Model (LLM), Student Model, assessment engine, and Q-Learning policy. CPL–CPMK–Sub-CPMK–indicator–rubric mappings provide assessment traceability, while the LLM is constrained by competency targets, difficulty, item type, and structured output requirements. The Student Model stores mastery and confidence estimates for each competency. Q-Learning acts as a sequential decision mechanism that selects the next assessment action, such as target competency, difficulty, and item type, rather than directly estimating student ability. The research follows a Research and Development approach covering requirements analysis, architecture design, prototype development, instrument validation, and an experimental evaluation plan against non-adaptive baselines. The resulting design establishes a closed loop from learner response to AI-supported evaluation, Student Model updating, reward calculation, and next-item selection. No empirical performance values are fabricated; quantitative claims are reserved for the planned validation study. The main contribution is a traceable adaptive-assessment architecture that combines OBE, rubric-constrained LLM assessment, executable code testing, and Q-Learning-based sequencing.