Traffic congestion is a crucial problem in urban environments with a significant impact on safety, social, and economic aspects. In this study, we discuss how accidents and traffic congestion are interconnected. For this purpose, we proposed a novel labeling approach to define congestion states based on accident variables. Three labeling techniques are introduced: formula-based, hotspot-based, and hybrid approaches. We focus on a hybrid approach, using the DBSCAN algorithm to identify the accident hotspots as clusters. Each cluster is assigned a congestion index score, which is categorized into congestion states (low, medium, and high). For deeper congestion analysis, we investigate two alternative models. Preliminary results show that the hybrid approach combined with the proposed Bayesian Network(BN) outperforms the other labeling approaches. Moreover, the results indicate the robustness of the Hybrid approach in traffic congestion analysis. The labels obtained from the hybrid approach are used to evaluate the BN model’s performance in comparison to five popular Machine learning (ML) models. The BN model outperformed all five ML models in accuracy, precision, recall, and F1 Score. Furthermore, the BN model was used for root cause analysis of accidents, assessing their likelihood of causing congestion. Our study indicates that scenario variables like local authority (city) and road category (rural, urban, or highway) significantly impact the congestion probability, which is of value for improving traffic management.