The document summarizes research on ranking the susceptibility of electrical feeders in the New York City power grid to failure using machine learning techniques. It finds that Rankboost and SVM performed comparably well in predicting failures over 7-day periods, while MartiRank performed slightly worse. All algorithms detected periods of concept drift where less historical data was optimal for prediction. Future work could integrate analysis of feeder subcomponents, better handle concept drift, and estimate time to failure.