Application of machine learning methods for identification and detection of extreme counterparties

The article explores the k-means clustering algorithm, for identifying extreme counterparties that pose an increased risk to business operations. Special attention is paid to analyzing the geographical localization of counterparties, enabling the detection of spatial anomalies such as mass registration at a single address or concentration in high-risk regions. The methodology of using k-means to segment counterparties into homogeneous groups and subsequently detect outliers based on distances to cluster centroids is described. The silhouette coefficient is employed to assess clustering quality, providing a quantitative measure of cluster compactness and the degree of isolation of anomalous objects. The results indicate that the proposed method reduces operational costs for counterparty verification and minimizes financial and reputational risks.