This week’s list of data news highlights covers May 8, 2021 – May 14, 2021 and includes articles about creating a national dashboard for rent debt in the United States and improving soccer strategy with AI.
The U.S. Department of Defense (DOD) is automating its financial management systems using machine learning. When DOD officials misclassify or mismatch transactions, humans have to manually correct records in the system’s spreadsheets, which can take two hours for every unmatched data point. By using machine learning algorithms to find and fix irregularities, DOD has been able to automate the pairing of unmatched transactions in its databases, saving time and money.
PolicyLink, a research and policy institute in California focused on racial and economic equity, has collaborated with the University of Southern California (USC) to create a national data dashboard for rent debt in the United States. The dashboard includes data on the number of households behind on rent, the estimated total rent debt, and the estimated rent debt per household at a state and county level. It shows that 53,000 households in Minnesota are behind on rent, 60 percent of which are households belonging to people of color.
Researchers at the University of California San Diego have developed an AI navigation system that helps robots in emergency departments deliver supplies and materials. When a patient’s condition deteriorates, doctors gather around them to deliver care. To help robots navigate these clustered environments efficiently, researchers trained an algorithm to navigate around them using more than 700 videos from YouTube. When tested against other robotic navigation systems, the tool generated the most efficient and safest paths.
Researchers at the U.S. National Oceanic and Atmospheric Administration (NOAA) and National Weather Service have used a supercomputer to develop a model that can predict the hourly threat of rip currents, which are strong ocean currents that kill more than 100 people each year in the United States. The model can predict the hourly probability for a rip current up to six days in advance, which lifeguards can use to forecast warnings to beach-goers.
The Pacific Northwest National Laboratory has partnered with the University of Washington to develop an app that uses smart sensors and machine learning to predict parking space availability in Seattle, Washington. The researchers trained the app’s machine learning model to predict availability up to 30 minutes in advance using parking behavior data collected from 300 sensors within 74 parking zones. The tool uses data that is updated every 10 to 15 seconds and provides a range of vehicle sizes that a space can accommodate.
Researchers at NASA have collaborated with the United Nations Food and Agricultural Organization (FAO) to use satellite data on soil moisture and composition to predict the breeding sites of desert locusts, which are a species of grasshoppers in East Africa that eat crops. With the satellite data, researchers have been able to create a data dashboard to inform farmers on when and where to spray pesticides and insect growth regulators.
Researchers and clinicians at the Peking University People’s Hospital in China have developed exhaled breath tests that use machine learning to detect lung cancer in patients. The team collected exhaled breath samples from 428 patients, 139 of which had lung cancer, and used ionization techniques and machine learning to identify molecules and chemical compounds indicative of cancer in each sample.
The U.K’s Food Standards Agency (FSA), a government department focused on food safety, has developed a model that can predict which imported foods should be sampled for testing. The team noted seasonal variations in the amounts of risky food items entering the country, indicating weather conditions have an impact on food risks. To better manage risks, the team has built a model to classify imported food and feed products that take into account the type of food import, the country of origin, and the weather in that country.
Researchers from the Massachusetts General Hospital have collaborated with the Massachusetts Institute of Technology to develop a tool that measures unconsciousness in patients under anesthesia. The team trained an algorithm to predict consciousness and unconsciousness under anesthesia using 33,000 recordings of brain electrical activity. When tested on 27 patients receiving surgery, the tool was able to detect a patient’s decreasing level of unconsciousness before the attending anesthesiologist did.
AI company DeepMind has partnered with Liverpool soccer club to analyze soccer and improve strategy. Using data on players collected from sensors, GPS trackers, and computer vision algorithms, researchers have shown AI models can predict how a specific team and lineup will react in a particular situation. For instance, using data on 12,000 penalty kicks taken across Europe in the last few seasons, the model found that strikers are more likely to aim for the bottom-left corner of a goal than midfielders.
Image credit: Wikimedia Commons