Comparison of Machine Learning Methods for Predicting Electrical Energy Consumption

This research investigates how to accurately predict electrical energy consumption to address growing global energy demands. The study employs three Machine Learning (ML) models: k-Nearest Neighbors (KNN), Random Forest (RF), and CatBoost. To enhance prediction accuracy, the researchers included a data pre-processing step using min-max normalization. The analysis utilized a dataset containing 52,416 records of power consumption from Tetouan City. The dataset was divided into training and testing sets using different ratios (90:10, 80:20, 50:50) to evaluate model performance. Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) were used to assess prediction accuracy. Min-max normalization significantly improved KNN's performance (reduced RMSE and MAPE). RF achieved similar accuracy with and without normalization. CatBoost also demonstrated stable performance regardless of normalization. Data pre-processing, specifically min-max normalization, is crucial for improving the accuracy of distance-based algorithms like KNN. Decision tree-based algorithms like RF and CatBoost are less sensitive to data normalization. These findings emphasize the importance of selecting appropriate pre-processing techniques to optimize energy consumption prediction models, which can contribute to better energy management strategies.