Combining Machine Learning and Dimensionality Reduction Techniques to Provide an Efficient Customer Segmentation: A Case of CAS Tecnologia Smart Grid Power System in Brazil

In the era of big data analytics, customer segmentation requires new approaches to analyze consumption behavior and develop targeted services. In this study, it was explored the K-medoids clusterization method with various similarity metrics and clustering scores, combined with PCA, t-SNE, and UMAP dimensionality reduction techniques to investigate typical and peculiar patterns in electricity consumption from a smart grid system in Brazil. By utilizing traditional and novel features and conducting features interpretation and pattern shift analysis, it was demonstrated the effectiveness of this framework in supporting decision-making processes for electric power companies and promoting sustainable energy management through customer segmentation.