Approximately 6.1 million individuals dropped off the food stamp rolls since President Donald Trump’s first full month in office in February 2017, according to the latest data from the U.S. Department of Agriculture (USDA).

The USDA data showed that 6,074,074 individuals discontinued their participation in the Supplemental Nutrition Assistance Program (SNAP) between February 2017— when the president completed his first full month in office— and November 2019.

Household participation in SNAP declined as well, with 2,489,315 households discontinuing SNAP.

There are currently 36,223,717 individuals and 18,448,588 households that are participating in SNAP.

When Trump took office, 42,297,791 individuals and 20,937,903 households were enrolled in SNAP.

Trump recently made it a point in his annual State of the Union address to stress that he helped the poor move off welfare to find jobs with his recent enactment of work requirements.

Under these work requirements, which had been enacted at the state level during the Obama years, those between the ages of 18-49 and without children or dependents who receive food stamps for more than three months in a 36-month period must work, go to school, receive job training, or volunteer to receive benefits.

Another way the USDA has been trying to keep enrollment in the food stamp program down is through the use of data-mining practices to identify food stamp fraud.

According to a report from the Government Accountability Office (GAO), computer algorithms went through SNAP purchase data in seven states and matched it up with retailer and eligibility data to see if there was fraud.

In Mississippi, the state reported $2 million in SNAP overpayments since the state started incorporating data-mining into its fraud detection efforts.

But the study was limited, as all seven states said high costs, data limitations, and organizational support affected their ability to use more advanced artificial intelligence-gathering techniques to ferret out fraud.

Other states, however, have done similar things with analytics for different welfare programs at the state level. Utah became one of the first states to modernize its unemployment insurance analytics system in 2006 after spending $14 million to overhaul it. In 2015, the state had one of the lowest fraud rates in the country at 1.3 percent.