Parameters Sensitivity Analysis of Ant Colony Based Clustering: Application for Student Grouping in Collaborative Learning Environment

Clustering analysis is one of the data analysis techniques that organizes items into clusters according to their degrees of similarities. In this context, bio-inspired algorithms have found success in solving clustering problems. Inspired by nature, Ant Colony based Clustering arises from ant colony behavior in organizing nests and clustering ants corpses. Accordingly, several researchers proposed different clustering algorithms that mimic the real ants behavior in forming cemeteries. However, the performance of a given algorithm depends strongly on its parameters settings. Indeed, it holds a large number of adjustable parameters that need to be instantiated by suitable values. In this paper, we study the parameters influence, more precisely the parameter
$\alpha $
which is responsible for adjusting similarity between objects. In fact, we analyze the impact of
$\alpha $
values on the performance of some well known Ant Colony based Clustering Algorithms applied to constructing team-works in a collaborative learning environment. After various bench tests, the choice of
$\alpha $
value is determined based on the best algorithm accuracy for each learning data-set. The experimental results prove that Ant Colony algorithms performance strongly depends on
$\alpha $
, especially when applied to large data-sets size. However,
$\alpha $
has a negligible influence on the algorithm’s accuracy when applied to small data-sets size. Obviously, the feature selection step could be ignored since it has a negligible influence on the algorithm performance even with different values of
$\alpha $
.