What happened
From 2015, EPA worked with the University of Chicago Energy & Environment Lab to develop a machine-learning model. The model forecasts which large quantity generators of hazardous waste are likely to violate RCRA rules. EPA writes that it is based on 15 years of historical data from across EPA programs. From fiscal years 2017 to 2019, EPA coordinated its regions to field test inspection targeting with the model. In its list of program evaluations for fiscal year 2020, EPA reports that a randomized evaluation compared the model with status quo targeting. It states that the model yielded a substantial increase in efficiency of detecting violations, with no increase in operating cost. EPA calls these results preliminary. It writes that the University of Chicago would release a working paper on how the model nearly doubled the violation detection rate. On its own page, the Lab reports an 82 percent higher detection rate. EPA then released predictive scores to its regions and to states for inspection targeting in fiscal year 2021. EPA is the deployer and a co-developer of the model.
Sources
-
Evaluation report, by an evaluator that the makers or users commissioned.
U.S. Environmental Protection Agency. FY 2020 EPA Program Evaluations, Office of Enforcement and Compliance Assurance (OECA), Activity 1: RCRA Data Analytics Evaluation, p. 1. Evaluation run by EPA with the University of Chicago Energy & Environment Lab.
The randomized evaluation completed in FY 2020 benchmarked the ability of the model to identify facilities with severe RCRA violations detected during inspection relative to status quo practices across Regions. Preliminary results demonstrate that the model yields a substantial increase in efficiency of detecting hazardous waste violations compared to current targeting practices, with no increase in operational cost.
-
Other, by the makers.
University of Chicago Energy & Environment Lab. AI and Inspection Targeting. Web page, read on 31 October 2026.
The field experiment showed that the model had an 82 percent higher detection rate than the EPA’s status-quo practices, with no increase in cost.
Details
AI system
- Workflow
-
The model was based on 15 years of historical data from across EPA programs. EPA regions inspected facilities chosen with the model and with status quo practices in a randomized comparison. Inspectors used the scores together with their own experience, skills and knowledge.
Follow-up
- Conditions for replication
-
Inspectors must combine the scores with their own judgment, as EPA describes. The model must be refined with new scores each year, which EPA planned with the Lab's support.
Entry details
- Sector
-
Environmental enforcement (hazardous waste).
- Purpose
-
The University of Chicago Energy & Environment Lab built the model with EPA to forecast which large quantity generators of hazardous waste are likely to violate RCRA rules, so that inspectors can focus on them.
30 more questions not yet checked
model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.
Cite this record
Peter H. Vartanian. (2026). RCRA predictive targeting model spots violations more efficiently (AIBD-1035). Artificial Intelligence Benefits Database, edition of 1 November 2026. https://petervartanian.xyz/aibd/records/AIBD-1035.html (archived at https://doi.org/10.7910/DVN/AADSII, version 7.0)