Predicting Power Consumption Anomaly Using Statistical and Supervised Machine Learning Techniques

Power has become an essential element of daily life in the modern world. At the same time, over usage of electricity may lead to excessive power consumption, causing the device to short circuit or be on fire. Therefore, it is crucial to monitor and forecast the anomalies in power consumption to avoid any tragedy. In this paper, the authors proposed a method of predicting anomalies in power consumption. The proposed method uses a statistical approach in labeling; the labeled power consumption data are then used to form the data instances. Later, supervised machine learning classification techniques, namely Support Vector Machine, Decision Tree, and Random Forest, are implemented on the data instances to predict the power consumption anomalies. The experimental results demonstrate that the precision, recall, and F1 score are achieving better results when the training dataset is larger. Besides that, the results show that the techniques used to handle the imbalanced data will affect the performance of models