An Efficient MCD-OSVM Model for Outlier Detection in IoT-Based Smart Energy Management Systems

Abstract

As Information, Communication, and Sensor Technologies (ICST) continue to evolve, data-driven innovations like the Internet of Things (IoT) and Smart Technologies, including Smart Energy Management Systems (SEMS), have become increasingly prevalent worldwide. Ensuring data quality is crucial for the effective implementation of IoT-based SEMS, as poor data management in these critical systems can significantly impact the quality of life for millions and potentially lead to severe disruptions and damage at a national level. In this research, an efficient One-class Support Vector Machine (OSVM) model is developed by deploying the Minimum Covariance Determinant (MCD) model at the data pre-processing phase to clean the training data This allow a better trained OSVM model that can be used for the outlier detection. The comparison between the efficient MCD-OSVM model and the base OSVM model, both based on the same original model, highlights a key difference in the training phase: the proposed model was trained with cleaned data using the MCD method, while the base OSVM model used the original, uncleaned data. Cleaning the dataset with an efficient method such as MCD improves the accuracy of OSVM model, an increase of 13.21% in average accuracy, while only increase the operation time 9.5 seconds, although the overall operation time can be further reduced as it is also found a cleaner training dataset will indirectly improve the execution time of OSVM models by allowing it to run on a lower NU parameter value.