Application of a Kernel Method Autoencoder-Based Clustering Approach Combining DBSCAN and K-means for Rail Transit Station Classification

With the rapid development of rail transit, station management faces increasing complexity and an explosive growth in data volume. To address the challenges posed by the high dimensionality and diversity of passenger flow data, this study proposes an enhanced clustering model that integrates KAE (Kernel method-based Autoencoder, KAE) for dimensionality reduction with a hybrid clustering approach combining DBSCAN and K-Means. The KAE is utilized to reduce the dimensionality of the high-dimensional data while retaining key features, which improves the effectiveness of the subsequent clustering process. In this model, DBSCAN is applied to identify high-density regions in the data, which are then used to initialize the K-Means algorithm. This ensures that the initial centroids are positioned in areas with high data density. The proposed approach combines the capability of DBSCAN to detect clusters of arbitrary shapes based on density with the ability of K-Means to refine cluster centroids for better convergence. The use of KAE further enhances the clustering performance by optimizing the feature space, making it more suitable for clustering. Experiments conducted on both rail transit and general datasets demonstrate that the proposed model achieves superior clustering performance compared to traditional methods. The results show improved classification accuracy and robustness, confirming the model's strong generalization capability across different types of data. This method provides an efficient and scalable solution for the classification of rail transit stations and offers new insights for managing and analyzing high-dimensional data in various applications.