Cost-Effective Feature Selection and Ordering for Personalized Energy Estimates

Background Selecting homes with energy-efficient infrastructure is important for renters, because infrastructure influences energy consumption more than in-home behavior. Personalized energy estimates can guide prospective tenants toward energy-efficient homes, but this information is not readily available. Utility estimates are not typically offered to house-hunters, and existing technologies like carbon calculators require users to answer prohibitively many questions that may require considerable research to answer. Aim We want to provide prospective tenants with personalized predictions for energy consumption that are certain (have narrow prediction intervals), accurate (are close to the true values), and cheap (require minimal burden on the user from answering questions). Giving energy estimates for a household requires eliciting information about that household; our goal is to strategically order questions so that the most informative features are answered first, to give confident predictions with minimal user burden. Data The Residential Energy Consumption Survey (RECS) records energy consumption by fuel type and around 500 household features for homes across the U.S.; we can use this dataset to learn relationships between household features and energy consumption to predict energy usage for prospective tenants. We restricted our analysis to the 2470 homes in the same climate zone as Pittsburgh. Methods For the task of providing personalized utility estimates to prospective tenants, we present a cost-based model for feature selection at training time, where all features are available and costs assigned to each feature reflect the difficulty of acquisition. At test time, we have immediate access to some features but others are difficult to acquire (costly). In this limited-information setting, we strategically order questions we ask each user, tailored to previous information provided, to give the most certain predictions while minimizing the cost to users. Results During the critical first 10 questions that our approach selects, prediction accuracy improves proportionally to fixed order approaches, but prediction certainty is higher. We can then make useful predictions at only 30% of the cost of the full-feature model. Conclusion Dynamically ordering questions at test time allows prospective tenants to receive certain and accurate estimates for energy consumption in potential future homes, without requiring them to collect prohibitively costly amounts of detailed information about each unit.