An Investigation on the Impact of Optimal Grouping and Regrouping on Learners’ Performance and Communicative skills in Collaborative Learning

Collaborative learning has emerged as a central pedagogical paradigm in higher education, fostering knowledge co-construction, peer support, and the development of interpersonal skills. However, its effectiveness is often constrained by challenges in group composition. Static grouping strategies, which assume that group membership remains fixed, frequently lead to unequal participation, unbalanced skill distribution, and reduced learner engagement. To address these limitations, this doctoral research develops and evaluates an artificial intelligence (AI)-enabled multi-objective optimization framework for adaptive group formation and regrouping in collaborative learning environments.The research introduces a novel framework grounded in metaheuristic optimization methods, specifically the Non-dominated Sorting Genetic Algorithm II (NSGA-II), to manage the complex problem of simultaneously balancing inter-group homogeneity and intra-group heterogeneity. The framework integrates multiple learner attributes, including engagement, communication, motivation, and performance-related skills, into a dynamic grouping process that operates iteratively across the learning cycle. Unlike traditional static methods, the system continuously monitors learner performance, updates individual profiles in real time, and adjusts group composition adaptively through feedback-driven regrouping. This approach ensures that group dynamics remain aligned with learners’ evolving competencies and promotes equitable opportunities for participation and skill development.A mixed-methods research design was employed to evaluate the framework. The empirical component was conducted with 55 undergraduate students enrolled in a course at the University of Portsmouth, UK. Participants engaged in collaborative tasks over six teaching weeks, with three contact hours per week dedicated to structured group activities. Performance was monitored weekly using a validated rubric across the four core dimensions of engagement, communication, motivation, and overall proficiency. Groups were initially formed through random allocation, then optimized using the proposed algorithm, and subsequently restructured at predefined checkpoints. Both formative and summative assessments, complemented by detailed qualitative observations, were employed to capture the evolution of learner performance and group processes.The findings demonstrate significant improvements across all four evaluated dimensions, with particularly strong gains in engagement and communication. Notably, low-performance bands were eliminated after the first regrouping, and a consistent upward trend was observed in learner outcomes across successive iterations. Statistical analysis confirmed that the intervention raised median scores, reduced disparities across the cohort, and fostered progression into higher performance bands. Qualitative evidence further illustrated how adaptive grouping supported equitable participation, enhanced motivation, and facilitated peer-to-peer learning through diversified group interactions.This thesis contributes both methodologically and pedagogically. Methodologically, it operationalises multi-objective optimization within a real educational context, providing a transparent and replicable framework for adaptive group formation. Pedagogically, it demonstrates how AI-enabled grouping strategies can enhance equity, inclusivity, and collaborative effectiveness in higher education. The work advances the field of Computer-Supported Collaborative Learning (CSCL) by aligning computational rigor with social constructivist principles, thereby establishing a bridge between algorithmic design and educational practice.The implications extend to instructors, programme leaders, and policymakers. For educators, the framework offers practical guidelines for aligning task design, assessment, and group formation in ways that foster dynamic and inclusive collaboration. For institutions, it provides a scalable model adaptable to larger cohorts, online delivery, and diverse disciplinary contexts. Future research directions include extending the algorithm to account for cognitive and metacognitive attributes, investigating long-term impacts on learner autonomy, and integrating the framework into mainstream learning management systems.This study demonstrates that adaptive, AI-enabled group formation and regrouping can serve as a transformative mechanism for enhancing collaborative learning, offering both theoretical contributions and actionable solutions to improve learning outcomes in higher education.