CNN-GRU-Based Feature Extraction Model of Multivariate Time-Series Data for Regional Clustering

Clustering-related research on data with time continuity is largely done through statistical analysis and thus does not fully reflect the data’s features. In this paper, we propose a CNN-GRU-based model to extract each variable’s time-dependent changes and features in multivariate data. We have utilized CNN to identify the features of each variable and derive trends over time based on GRU. Fuzzy C-means clustering is performed based on this feature and overlapped cluster results are finally obtained. Experiments were conducted using two years of card usage data to extract the features according to the local consumption industries and apply these to regional clustering. The proposed method’s performance is evaluated by comparing the proposed method with data characterization and clustering methods used in existing research.