Performance evaluation method for different clustering techniques

Abstract

This study presents a comprehensive methodology to objectively evaluate various clustering techniques applied to electrical demand profiles (EDPs). The effectiveness of Self-Organizing Maps (SOM), Fuzzy C-Means (FCM), and Hierarchical Clustering (HC) is analyzed, revealing that these methods achieve anomalous value percentages below 17.1%. The proposed approach includes a statistical framework based on confidence intervals to classify data as typical or atypical, thereby facilitating the selection of the most appropriate clustering technique based on the characteristics of the dataset. To evaluate the methodology, an analysis of probability distributions is used, comparing it with three internal validation techniques through the implementation of specific criteria. These metrics provide insight into the dispersion and distribution of the EDPs, allowing for a robust evaluation of how variations in data impact clustering outcomes. The results indicate that the SOM, FCM and HC techniques exhibit strong adaptability to different patterns of variability, making them suitable for diverse applications in energy management. This research contributes valuable tools for optimizing the classification of EDPs, enhancing the understanding of consumption behaviors in the electricity sector.