Feature Selection for Electrical Demand Forecasting and Analysis of Pearson Coefficient

Demand forecasting contributes to the stable operation of the electrical grid. Increasing computational complexity with the emerging big data in the interconnected electricity grid poses a challenge on the effective electrical demand forecasting. The emerging machine learning algorithms are capable of handling this big data, but their performance depends on the quality of the input features in the dataset. Feature selection filters relevant and informative attributes, improving the quality of the input dataset hence facilitating better prediction. In this article, the feature selection for electrical demand forecasting is presented. The relationship of input features with the electrical demand and relation between the input features are analysed. Correlation coefficient, variance inflation factor (VIF) and analysis of variance (ANOVA) based techniques are examined to identify the relevant attributes. Performance of ANOVA is analysed via sum of squares (SS), mean square (MS) and degree of freedom (df). A brief discussion on the key issues concerning multicollinearity and the application of Pearson coefficient is also presented. Challenges that require further attention in feature selection research are discussed at the end.