My paper investigated non-technical loss (NTL) detection at a North American utility. Historically, finding NTL for an electric utility was a labor-intensive process that required randomly visiting properties and checking the meter’s operation. With the introduction of smart meters in utility operations there is now meter event data and interval consumption data which can be used in NTL analytics to better identify which properties likely have NTL.  

Using this meter and consumption data, and gradient boosted decision trees (GBDT), the paper reported the findings for improving the efficiency of a North American’s NTL lead pipeline by ranking the existing leads. The GBDT models used were an attempt to replicate an NTL research paper identified through the literature review.