Machine Learning and Forecasting: A Review

Interpretable Short-Term Electrical Load Forecasting Scheme Using Cubist

“…The lower the MAPE and CVRMSE values, the better the forecasting model's predictive performance. However, it is known that the MAPE and CVRMSE increase significantly when the actual value tends to zero [ 35 , 52 ]. The MAPE and CVRMSE are calculated using ( 10 ) and ( 11 ), respectively, where y t and are the actual and forecasted values at time t, respectively, is an average of the actual values, and n is the number of observations.…”

“…When constructing a forecasting model, datasets are usually divided into a training set and a test set, and then the forecasting model is built using the training set and verified using the test set [ 35 ]. However, conventional time-series forecasting model evaluation could exhibit unsatisfactory prediction performance when there is a significant gap between the training set period and the test set period.…”

“…Despite these efforts, there are still limitations. For instance, when a forecasting model based on these methods is constructed using conventional time-series forecasting model evaluation [ 35 ], we can determine the essential input variables by analyzing the variable importance in the model constructed from the training dataset. However, conventional time-series forecasting model evaluation performs unsatisfactorily when there is a significant gap between the training set period and the test set period [ 14 ].…”