Improved Energy Consumption Prediction using XGBoost with Hyperparameter tuning

Abstract

There is a strong need for energy consumption predictions as it is growing rapidly year by year. These forecasts are beneficial for power production and supply companies and even for the country. Although energy is not the only input that determines the level of production and the degree of economic development of a country, it is highly important for economic growth. It is only by consuming a certain amount of energy that countries can achieve a certain level of economic growth. Hence, it is highly significant to predict energy consumption as it is a growth indicator. Machine learning approaches can forecast the future based on past customer energy consumption as well as various other characteristics. As there are a large number of features that affect the hourly energy consumption, this paper proposes a system that mainly uses the extreme gradient boosting algorithm in the analysis and predictions of energy consumption with feature selection and hyperparameter tuning, achieving the results of hourly energy prediction with a relative error of 7.76% and RMSE of 3.31 kWh.