Speaker
Description
This paper examines the applicability of machine learning techniques to spatial data modeling and the identification of urban patterns using a limited set of demographic and geographic indicators. Drawing on sector-level data for Bucharest obtained from the National Institute of Statistics of Romania, the analysis incorporates variables such as surface area, total population, population density, and distance to the city center. The latter is computed as the Euclidean distance between sector centroids and a predefined central reference point, ensuring a consistent spatial measure across all observations. The dataset is processed through a standardized and reproducible analytical workflow that includes data preprocessing steps such as feature normalization to ensure comparability across variables with different scales. Clustering is performed using the k-means algorithm, with the optimal number of clusters determined through the elbow method, allowing for an informed balance between model simplicity and explanatory power. To further support interpretation and visualization, principal component analysis (PCA) is applied, reducing dimensionality while preserving the most relevant variance in the data. The findings reveal three distinct spatial groupings that correspond to peripheral, intermediate, and central urban structures. Sector 1 emerges as a clearly differentiated cluster, characterized by a relatively large surface area, low population density, and greater distance from the city center. In contrast, Sectors 2, 4, and 5 exhibit more compact spatial configurations and are more centrally located, reflecting higher levels of urban concentration. Sectors 3 and 6 display intermediate profiles, combining moderate spatial extent with balanced demographic characteristics. The silhouette score of 0.266 indicates moderate cluster separation, which is consistent with the limited size and dimensionality of the dataset. Despite these constraints, the results highlight the potential of relatively simple machine learning approaches to support exploratory spatial analysis. Overall, the study demonstrates that interpretable and reproducible workflows can generate meaningful insights into urban spatial structure and provide a solid foundation for the development of more advanced modeling approaches in future research.
