The advancement in artificial intelligence technology has greatly enhanced load forecasting precision. Nonetheless,certain load curves within the training set experience low forecasting accuracy.This is attributed to data imbalance resulting from the load curves belonging to a limited number ofclasses.Consequently,the model fails to learn sufficiently leading to compromised forecasting accuracy.To address the issue of reduced prediction accuracy due to data imbalance in electric load forecasting,a solution involving classification before load prediction is proposed.The method involves clustering historically similar load profiles using an improved K-Medoids approach,constructing classification labels,and feature sets for electric load.Subsequently,classification is performed based on temporal features,and the RUSBoost algorithm is employed to effectively address data imbalance issues during the classification process.Finally, the LightGBM algorithm is utilized for load predictionwithin each class.Experiments on public datasets demonstrate the effectiveness of the proposed method in forecasting certain types of loads.The MAPE is 2.95%,and the RMSE is 175.71 MW.For the forecasting of conventional loads,the MAPE is 3.52%,and the RMSE metric is 195.84 MW, which is relative to other methods that also have better forecasting effect.