reillymj writes "Every so often, a 300-pound manhole cover blows sky high in Gotham, followed sometimes by a column of flame and smoke. (There are a few hundred 'manhole incidents' per year in the city, not all of them this dramatic.) Researchers from Columbia University applied machine learning algorithms to Con Edison's warren of aging electrical wires and sewage access points around Manhattan. As the system learns where dangerous mixtures of sewer gas and decrepit wiring are likely to come in contact, it makes forecasts about trouble spots, including where the next explosion may occur. The team has just completed rankings for manholes in Brooklyn and the Bronx, and plans to return to Manhattan's grid, armed with the most recent inspection and repair data." The research was published in the July issue of Machine Learning.