New research suggests machine learning could help officials more effectively monitor potential violators of environmental regulations.
 |
| Machine-learning could help environmental regulators identify which facilities are most likely to fail an inspection. Photo by EPA-EFE/RYAN TONG |
By Brooks Hays, UPI
Regulatory agencies tasked with protecting environmental and public health are regularly understaffed and underfunded, but new research suggests machine learning could help officials more effectively monitor potential violators.
The Environmental Protection Agency and partnering state agencies are responsible for monitoring the regulatory compliance of 300,000 facilities. Regulators, however, only have the resources to inspect less than 10 percent of those facilities each year.
To help the EPA catch violations, student researchers at Stanford University designed a model to identify facilities most likely to fail an inspection. Scientists trained the machine learning to interpret a variety of risk factors, including the facility's location, industry and inspection history.
The machine learning model can assess risk scores based on factors linked with previous violations. Scientists used the model to predict how many violations regulators would find if they prioritized inspection by the model's risk assessments.
Under a scenario unconstrained by resource limitations, the models showed regulators could catch up to seven times more violations. Under a more realistic scenario, with budgetary constraints, the machine learning-powered risk scores could help regulators identify twice as many regulatory violations.
Just like human regulators, however, computer models aren't perfect.
"They can perpetuate bias at times and they can be gamed," Stanford grad student Miyuki Hino said in a news release.
If a facility operator could figure out how the algorithm worked, they could potentially manipulate data to ensure the facility would be assessed a low risk score. Operators could also alter management practices, beefing up compliance in anticipation of an inspection and relaxing standards when the risk of an inspection was low.
Scientists acknowledged the model fails to account for changes in regulatory priorities and environmental protection technologies. However, the model's creators said updates to the algorithm could help ensure its long-term efficacy.
Student researchers described the model's potential benefits and flaws in a paper published this week in the journal Nature Sustainability.
"This model is a starting point that could be augmented with greater detail on the costs and benefits of different inspections, violations and enforcement responses," said Stanford graduate student Nina Brooks.
Regulatory agencies tasked with protecting environmental and public health are regularly understaffed and underfunded, but new research suggests machine learning could help officials more effectively monitor potential violators.
The Environmental Protection Agency and partnering state agencies are responsible for monitoring the regulatory compliance of 300,000 facilities. Regulators, however, only have the resources to inspect less than 10 percent of those facilities each year.
To help the EPA catch violations, student researchers at Stanford University designed a model to identify facilities most likely to fail an inspection. Scientists trained the machine learning to interpret a variety of risk factors, including the facility's location, industry and inspection history.
The machine learning model can assess risk scores based on factors linked with previous violations. Scientists used the model to predict how many violations regulators would find if they prioritized inspection by the model's risk assessments.
Under a scenario unconstrained by resource limitations, the models showed regulators could catch up to seven times more violations. Under a more realistic scenario, with budgetary constraints, the machine learning-powered risk scores could help regulators identify twice as many regulatory violations.
Just like human regulators, however, computer models aren't perfect.
"They can perpetuate bias at times and they can be gamed," Stanford grad student Miyuki Hino said in a news release.
If a facility operator could figure out how the algorithm worked, they could potentially manipulate data to ensure the facility would be assessed a low risk score. Operators could also alter management practices, beefing up compliance in anticipation of an inspection and relaxing standards when the risk of an inspection was low.
Scientists acknowledged the model fails to account for changes in regulatory priorities and environmental protection technologies. However, the model's creators said updates to the algorithm could help ensure its long-term efficacy.
Student researchers described the model's potential benefits and flaws in a paper published this week in the journal Nature Sustainability.
"This model is a starting point that could be augmented with greater detail on the costs and benefits of different inspections, violations and enforcement responses," said Stanford graduate student Nina Brooks.
COMMENTS