Predicting total household energy expenditures using ensemble learning methods


Kesriklioglu E., OKTAY E., KARAASLAN A.

ENERGY, vol.276, 2023 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 276
  • Publication Date: 2023
  • Doi Number: 10.1016/j.energy.2023.127581
  • Journal Name: ENERGY
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, PASCAL, Aerospace Database, Applied Science & Technology Source, Aquatic Science & Fisheries Abstracts (ASFA), CAB Abstracts, Communication Abstracts, Compendex, Computer & Applied Sciences, Environment Index, INSPEC, Metadex, Pollution Abstracts, Public Affairs Index, Veterinary Science Database, Civil Engineering Abstracts
  • Ataturk University Affiliated: Yes

Abstract

Total household energy expenditures are a complex topic because so many behavioral, technological, environ-mental, and policy variables can affect expenditures. This study aimed to develop a high-performance ensemble learning (EL) model to classify total household energy expenditures. For this purpose, household consumption data from 11,521 households were examined using the Household Budget Survey 2019 data set that the Turkish Statistical Institute (TURKSTAT) published. In addition to the variables directly related to household energy expenditures, new variables were created within the framework of the literature and under the guidance of expert opinion. The prepared data were passed through data preprocessing, modeling, prediction, and perfor-mance evaluation stages using the open source RapidMiner software program. Classification performances of machine learning and EL methods were compared. Aside from k-nearest neighbor, decision tree, naive Bayes, random forest, gradient boosted trees, and DFNN classifiers, the study used bagging, boosting, voting, and stacking EL methods. The stacking EL method in the ALL model and bagging EL method in the deep feed forward neural network (DFNN) classifiers achieved the highest performance among EL methods. The accuracy value of the stacking and bagging methods was 0.984. The results indicate that EL methods can enhance individual machine learning methods significantly.