Simplificação da visualização de mapas de processos na gestão pública municipal por meio de agrupamento

Abstract

Quality management in Brazilian municipal public administration faces chronic inefficiency challenges, perpetuating an empirical management model despite increasing digitalization and the vast electronic processing of procedural data. Process Mining emerges as an analytical solution, but its direct application in high-variability environments, such as municipalities, frequently results in the generation of unintelligible visual models, known as "spaghetti processes". Addressing this gap, this work conducts a comparative study among clustering algorithms (K-Means, Hierarchical, and DBSCAN) to evaluate their effectiveness in simplifying the visualization of administrative workflows using real data from a municipality. The methodology involved the extraction and preprocessing of a log with 167,175 events, the structural vectorization of the traces, the application of a topological filter retaining the 13% most frequent transitions, and the execution of the clustering algorithms. The evaluation was conducted using computational performance metrics, calculation of the Silhouette Coefficient, and qualitative visual inspection of the projected maps. The quantitative results revealed a clear trade-off : while K-Means and DBSCAN demonstrated high computational efficiency (memory consumption below 35 MB), K-Means failed in semantic cohesion (Silhouette score of -0.1695). DBSCAN stood out as a noise filter by isolating operational anomalies (Silhouette coefficient of 0.4255). In contrast, the Hierarchical Algorithm, despite requiring a high computational cost (peak of 8.4 GB of RAM), delivered a substantially superior structural quality (Silhouette Coefficient of 0.4406). It is concluded that clustering algorithms demonstrated effectiveness in simplifying process visualization in public management. Despite the trade-offs inherent to each model, the combined application of these techniques allows untangling data complexity, transforming chaotic flows into actionable visual tools for auditing and evidence-based decision-making.