Predicting Energy Customer Vulnerability Using Smart Meter Data

Ushakova, Anastasia and Jankin MIkhaylov, Slava (2018) Predicting Energy Customer Vulnerability Using Smart Meter Data. Working Paper. UNSPECIFIED.

Abstract

Supporting vulnerable consumers and reducing fuel poverty are major priorities for policy makers in the energy sector. With the availability of streaming data from smart meters we are able to develop simple and reliable methods of identifying vulnerable energy customers and as a result develop targeted policy interventions. This study investigates how vulnerable customers can be identified from natural gas consumption data. Neural networks, random forest, naive Bayes, and support vector machines were assessed for classification of consumer vulnerability. Random forest, with the prediction accuracy of 94.6 percent, outperforms other prediction models. Our study provides additional evidence that machine learning methods can be deployed by policymakers and insights teams to predict vulnerability from patterns of consumer behaviour.