A Comparative Analysis of Class Imbalance Handling Techniques for Deep Models in the Detection of Anomalies in Energy Consumption
Abstract
Description
<p>Detecting anomalies in energy consumption is critical for efficient energy management, fault detection, and<br>sustainability. However, the challenge of class imbalance, where normal consumption data vastly<br>outweighs anomalous instances, presents significant difficulties in building accurate predictive models.<br>This paper conducts a comparative analysis of class imbalance handling techniques for deep models in<br>detecting anomalies in energy consumption data. Specifically, controlled experiments are used to evaluate<br>the performance of deep learning models, such as convolution neural networks (CNN), long short-term<br>memory (LSTM) and BiLSTM deep algorithms as well as synthetic data generation (SMOTE), costsensitive learning, and generative adversarial networks (GAN) tailored to address the imbalance issue.<br>Through a comprehensive empirical study using a real-world energy dataset, we assess the models'<br>effectiveness based on area under the curve (AUC), precision, recall, F1-score, and their ability to<br>generalize across different levels of imbalance. This research contributes to improving model selection for<br>practitioners facing the class imbalance challenge in the energy sector. </p>
Keywords
Citation
Collections
Endorsement
Review
Supplemented By
Referenced By
Rights and licensing
info:eu-repo/semantics/openAccess
Creative Commons Attribution 4.0 International
https://creativecommons.org/licenses/by/4.0/legalcode
Creative Commons Attribution 4.0 International
https://creativecommons.org/licenses/by/4.0/legalcode