Time Series Data Splitting for Short-Term Load Forecasting

Inhalt:
This paper describes the implementation of two different approaches for data processing used in short-term electric load forecasting. For the prediction of the load two machine learning techniques were used. These are Linear Regression and Ensemble Decision Tree Regression. The real-life imperfect data were split into training and test sets using standard holdout method and proposed time series split method, which is the variation of time series cross validation. The suggested data splitting technique solves the overfitting issue and also yields to increase in performance for the otherwise overfitted machine learning model, however, also slightly decreases the accuracy of another model.