A Computer Adaptive Test Framework Leveraging Genetic Algorithm for Optimizing Testlet-Based Question Selection and Randomization

Abstract

Computer Adaptive Testing (CAT) has ushered in a new era in assessment by
providing a more efficient, accurate, and tailored testing experience for measuring
each test-takers' abilities. However, optimizing question selection and
randomization remains a significant challenge. In the framework of CAT lies the
question selection and randomization mechanism, which determines the next item
to be administered based on the test-taker's responses. Traditional approaches to
question selection often rely on Heuristic Randomization (HR), which can lead to
suboptimal item selection and compromised test validity. To address these
limitations, this study proposes a novel CAT framework that leverages Genetic
Algorithms (GA) to optimize testlet-based question selection and randomization.
The methodology adopted supports Object-Oriented Analysis and Design (OOAD)
using Agile Methodology. The proposed Genetic Algorithm Randomization (GAR)
employs a multi-parameter fitness function, incorporating question difficulty,
discrimination, time, and learning objective coverage. Through iterative evolution,
the algorithm identifies the optimal question set combination that maximizes test
efficiency. By harnessing evolutionary principles, GAR optimizes the selection of
CAT test questions. The results demonstrate that the possibility of optimizing CAT
with GAR for a better question selection and randomization process. The findings
of this research have significant implications for the development of more
sophisticated CAT systems, ultimately leading to improved assessment outcomes
and better-informed decision-making in education and professional certification.