<![CDATA[The increase in electricity demand in line with population growth and economic activity requires an accurate and reliable electricity consumption forecasting system. Short-term electricity consumption predictions are an important component in energy system planning and management, particularly to support grid stability and operational efficiency. This study aims to model electricity consumption predictions using a machine learning regression approach and analyze the factors that most influence electricity consumption based on historical data. The dataset used consists of smart meter data with a 30-minute time interval that has undergone data cleansing, data transformation, and feature engineering, including the formation of lag features and temporal features. Three regression algorithms were used, namely Linear Regression, Random Forest Regression, and Gradient Boosted Trees Regression. Model evaluation was performed using the Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R²) metrics. The results show that Linear Regression provides the best performance on the test data with an RMSE value of 0.156, MAE of 0.125, and R² of 0.140, and demonstrates stable generalization capabilities. The analysis of influencing factors reveals that historical consumption variables, particularly Avg_Past_Consumption and electricity consumption lag features, are dominant factors in the prediction, while environmental variables contribute relatively less. These findings provide practical implications for short-term energy demand planning by enabling more accurate load estimation and supporting data-driven decision-making through interpretable electricity consumption patterns.]]>
Copyrights © 2026
