Forecasting Electricity Consumption in Five Regions of Dhaka City: A Comparative Assessment among various ML-Boosting Models

Predicting electricity usage is vital to maintain a consistent power supply, making it a top concern for all nations. However, in Bangladesh, research in this area is inadequate. So, this research focuses on predicting power consumption in some regions of Dhaka City using XGBoost, LightGBM, and CatBoost algorithms. Monthly consumption, sanctioned load, meter type, and tariff are the key factors employed in the prediction process where the grid search technique is utilized to discover the most effective hyper-parameters. Model performance is evaluated by MAE, RMSE, and R2 scores. Experimental results indicate that CatBoost outperforms other algorithms (RMSE 2.0% to 42.0
^%
), although LightGBM offers faster training. Implementing these techniques holds significant promise for enhancing power infrastructure planning and energy forecasting, ultimately minimizing system losses and equipment damages.