Leveraging Interquartile Range and Isolation Forest for Abnormal Power Consumption Prediction

The progression of time has brought in new challenges, particularly in power consumption and regeneration, affecting individuals, cities, and nations globally. From ancient fire fueling to modern electricity, energy development is integral to global advancements. Considerations such as global power consumption, economic growth, comprehensive energy systems, and mitigating global warming are important. Therefore, avoiding overuse or power wastage is essential. Through data collection and analysis, energy usage patterns can be forecasted using machine learning techniques. In this research, the authors build a model to predict the unusual power consumption. The model comprises three stages. The first stage involves labeling power consumption as normal or abnormal using the Interquartile Range and Isolation Forest. The second stage addresses the imbalanced data from the first stage using SMOTE. The third stage employs machine learning classifiers to construct the prediction model. Experimental results demonstrate that applying Isolation Forest labeling, SMOTE oversampling, and utilizing Random Forest classifier shows better results in performance evaluation.