Exploring Energy Poverty Indicators Through Artificial Neural Networks

Abstract

The present paper is an important step in the development of energy poverty research, introducing artificial intelligence to the analysis of the problem. Literature has shown that conventional mathematical/statistical tools fail to take into account the complexity of different human responses to the energy problem. This weakness is attempted to be overcome with the use of artificial neural networks (ANNs), through the case of Greece. For the purposes of the research, a neural network, i.e., multilayer perceptron, of the machine learning application/tool “WEKA” was used and trained, in order to predict “objective” energy poverty based on “subjective” aspects. More precisely, three typical objective indicators of energy poverty were selected as output variables, namely 10%_actual (based on actual expenses), 10%_required (based on required expenses), and compression of energy needs (CEN), and five different subjective indicators were selected as input variables. The analysis showed that certain human behaviors/subjective indicators can predict objective energy poverty at a marginally satisfactory level, in the order of 56–58%. From the variety of human behaviors and responses, the restriction of other essentials in order to meet heating needs proves to be the key parameter of predicting energy poverty based on the indicator 10%_actual, while the condition of living in an inadequately heated home emerges as the key parameter reflecting energy poverty based on the CEN indicator. Artificial intelligence is expected to be a promising tool in understanding energy poverty and, therefore, in planning effective energy poverty policies.