Age Group Detection in Stochastic Gas Smart Meter Data Using Decision-Tree Learning

Abstract

Smart meters are the next generation gas and electricity meters where the meter readings are presented digitally and accurately to the consumer via an In-Home Display unit. Access to the data sets generated by smart meters is becoming increasingly prevalent. As such, this paper presents an approach for detecting age groups from aggregated smart meter data. The benefits of achieving this range from healthcare cluster mapping for smart resource allocation and intelligent forecasting, to anomaly detection within age-range groups. The technique proposed and presented in this paper uses a cloud analytics platform for the data processing. Using this approach, the classification is able to achieve a 75.1% AUC prediction accuracy using a two-class decision forest and a 74.6% AUC with a boosted two-class decision tree. A two-class linear regression model, which is able to achieve a 53.7% accuracy, is applied as a benchmark for comparison with the decision tree approach.