Artificial neural networks and linear regression prediction models for social housing allocation: Fuel Poverty Potential Risk Index

Abstract

Fuel poverty is a pertinent issue for vulnerable households both in industrialized and developing countries, which is related to energy prices and accessibility of energy services. This research explores the feasibility of predictive models to prevent fuel poverty through the Fuel Poverty Potential Risk Index (FPPRI). Two statistical models, multiple linear regression (MLR) and artificial neural networks (ANN), have been developed and applied to predict the probability of low-income households falling into fuel poverty when being allocated a social dwelling. The case study used to validate the model is located in the Bio-Bio Region of Chile and the households considered belong to the most vulnerable social strata. The models have considered the design and constructive features of common typologies of Chilean social dwellings, family income levels, changes in energy usage patterns and energy prices. Through extensive simulation and testing, ANNs have been found to be more accurate than MLRs for all situations, with a R2 coefficient above 99.6% and 80.7% respectively, despite their greater complexity. The result of this research can be useful in providing tools to fairly and accurately assign social dwellings to vulnerable households to prevent them from falling into fuel poverty.