Simple Approaches to Missing Data for Energy Forecasting Applications
Energy forecasting is not only a prominent sub-discipline in energetics, but also an arduous challenge to acquire electrical and climatological data that might have some missing values frequently due to power outages or equipment failures. There are several methods in the literature to treat missing data in datasets. Applying each method causes different accuracy of results in performance metrics. In this paper, a comparison of simple approaches named as linear interpolation method and marginal mean imputation method to missing data for energy forecasting applications using multilayer perceptron neural networks (MLPNN) and support vector machines (SVM) is presented through a case study of electrical energy consumption data and climatological data of a hospital in the Mediterranean Region.