Hybrid Bio-inspired Ant Colony Clustering approach for Constructing Collaborative Learning Teams

Abstract

Collaborative learning is an umbrella term for a wide variety of educational forms. It encourages students working in groups of two or more , to explore and share resources, knowledge, solutions, ideas, and thoughts to solve problems. Unlike traditional learning methods, which are based on individual learning performance Abid et al. (2024), collaborative learning helps learners at various performance levels work together in small groups toward a common goal. As a result, each member is accountable for both their peers’ learning and their own. Collaborative learning practitioners emphasize that this approach is fundamentally about building learning communities. It aims not only to improve learners’ academic skills but also to develop their interpersonal skills Asad and Qureshi (2025) Männistö et al. (2020). However, one of the key challenges in collaborative learning is creating groups where, learners need to feel safe, comfortable, when assigned to a suitable team. Therefore, forming the appropriate groups is one of the fundamental pillars of the collaborative learning. This issue has inspired many researchers in the educational field area, who apply different algorithms to address the grouping problem and discover optimal learner groups. Some researchers rely on computational approaches Chen and Li (2024), while others draw on heuristic ideas inspired by entomological studies of living organisms in nature Kiran et al. (2022). Nature-inspired meta-heuristic algorithms, developed based on principles drawn from biological evolution. They are well known in machine learning for addressing optimal solutions of complex problems. Numerous bio-inspired algorithms exist in the literature, such as Genetic Algorithms (GA) Forrest (1996), Particle Swarm Optimization (PSO) Kennedy and Eberhart (1995); Shami et al. (2022), Ant Colony Optimization (ACO) Dorigo and Di Caro (1999), Bee Colony Optimization (BCO) Teodorovi´c et al. (2022), etc. These algorithms have been successfully adapted and have found success in solving clustering and NP-complete problems. Among several bio-inspired collaborative systems, ant-based clustering techniques have proven successful in solving clustering problems. These methods are inspired by the ecological study of social ants. Being talking about social, means that ants cannot survive on their own since they belong to the social category. Therefore, they demonstrate remarkable coordination of activities among colony members. In recent years, ant-based clustering techniques received special attention from the research community due to their performance in exploratory data analysis across many fields, including collaborative robotics as in (Kallel et al. (2008), Chatty et al. (2011); Chatty et al. (2012); Chatty et al. (2013)). However, further investigation is needed to address issues related to performance, convergence, robustness, stability, etc. Our research journey revealed two main challenges. On one hand, bio-inspired algorithms involve different parameters which is significantly impact their performance. Parameter tuning (or setting) has attracted considerable research attention over the several researchers in the past decade Nannen and Eiben (2006) as even though the algorithm is efficient, setting inappropriate values of parameters may lead to a low-quality solutions. For example, in Hassanat et al. (2019), the authors stated that the integration among parameters such as mutation, crossover rates, and population size is vital for successful GA search. Similarly, the searching capability of the PSO algorithm is directly influenced by its three main control parameters (inertia weight, cognitive acceleration coefficient, and social acceleration coefficient). In fact, as noted by Carlisle and Dozier (2001), Trelea (2003) and Van den Bergh and Engelbrecht (2006), a priory tuning of PSO control parameters can significantly improve performance but remains highly sensitive to these settings. Likewise, ant-based clustering-inspired by larval sorting activities and corpses clustering observed in real ant colonies, the ant based clustering has emerged as a new method for solving clustering problems. Just like other bio-inspired algorithms, ant colony algorithms require an initial setting of parameters before starting. On the other hand, conventional ant colony–based clustering methods tend to leave certain data items unclustered or isolated. Since this thesis focuses on data clustering (or student team formation), one of the key challenges is ensuring a complete partition of the dataset without any remaining outliers. To tackle these issues, we propose a structured research approach comprising of four interrelated contributions. The first contribution establishes the theoretical and methodological basis through systematic literature review, which is necessary to move forward. The second contribution builds directly upon this foundation to select relevant attributes, enabling us to explore the selected educational datasets and their features in greater depth. The third contribution presents a comparative study of ant colony clustering parameters’ effects and their influence on small and large educational datasets. The final contribution leverages the findings from previous stages to propose a hybrid bio-inspired Ant Colony-based clustering algorithm integrated with the deterministic Kmeans algorithm (KM-AC) to form synergistic student groups based on academic and social attributes, thus offering an optimized solution to the research clustering problem outlined earlier. Each of these contributions is presented in a dedicated chapter: First, we present the different aspects of collaborative learning and identify the current state-of-the-art research on student group formation in learning environments Abid et al. (2016). Second, we analyze the impact of feature selection techniques on the classification task to identify relevant features that help improve the performance and effectiveness of our proposed approach Abid et al. (2017). Then, we study the parameters’ influence, more precisely on the parameter α, which is responsible for adjusting similarity between objects and its effect on small and large educational datasets Abid et al. (2023). Finally, we define our proposed hybrid bio-inspired Ant Colony clustering approach and analyze the influence of dataset size and group composition on clustering performance. In addition, we also analyze how dataset size affects algorithm performance Abid et al. (2025).