This paper considers methods for analyzing urban transportation networks based on their representation as weighted graphs reflecting travel-time characteristics. An approach to assessing the connectivity of street networks is proposed, which relies on constructing a proximity matrix using a Gaussian similarity function followed by graph-based clustering of transport links. The influence of similarity-function parameters and local scaling strategies on the properties of the resulting cluster structures, including their stability and interpretability, is analyzed. In contrast to traditional models employing a fixed smoothing parameter, a self-tuning Gaussian kernel is introduced, in which the scale parameter is defined individually for each vertex based on the characteristics of its local neighborhood. This approach enables adaptive edge weighting and allows the spatial heterogeneity of urban transportation networks to be taken into account. Functional zones of the city are identified using the Leiden algorithm, which optimizes the modularity of the cluster structure. A comparative analysis of clustering results obtained under different levels of traffic congestion is performed, reflecting changes in travel conditions from free-flow to highly congested regimes. The experimental results show that, compared with models using a fixed smoothing parameter, the proposed approach yields more stable cluster partitions of the transport graph and makes it possible to identify consistent patterns in the transformation of network connectivity as traffic conditions vary. Visualization of weight distributions and cluster structures confirms the sensitivity of the model to changes in traffic conditions. The proposed approach can be applied to the analysis of urban transport connectivity and the study of spatio-temporal patterns of urban mobility.