Dataset: Empirical study on the correlation among architectural smells

Abstract: Context. Architectural smells are indicators of architectural technical debt and may compromise the maintainability and evolvability of software systems. Hence, it is important to carefully study and monitor them. Objective. This study aims to investigate statistical collocations, i.e., correlations among architectural smells affecting the same components, to assess whether specific patterns of smells frequently appear together, forming clusters. Method. We constructed a large-scale dataset using an automated data collection pipeline with Arcan tool, analyzing 782 Java and Python projects. We applied PCA and Association Rule Mining to identify recurrent co-location patterns, finally illustrating some examples of patterns. Results. The analyses reveal possible co-location patterns: dependency-related smells (Cyclic Dependency, Hub-like Dependency, Unstable Dependency) often cluster with hierarchy-related smells (Cyclic Hierarchy, Wide Hierarchy). Other smells, such as God Component and Deep Hierarchy, are infrequent and do not significantly contribute to dominant patterns. ARM confirms these associations with high conviction but generally low support values. Conclusion. Architectural smells do not occur only in isolation but also in recurring clusters that point to deeper structural issues. These findings suggest that architectural degradation may arise from interaction between smells rather than individual ones, highlighting the need for analysis of smell collocations.