Methods, systems, and apparatus, including computer-readable storage devices, for tracking and analyzing service outages. Weather data and outage data are accessed. Based on the weather data and the outage data, weather conditions corresponding to service outages are identified. A machine learning system is trained to estimate training a machine learning system to estimate effects of weather conditions on utility systems. Data for a particular utility system is provided to the trained machine learning system, and in response, output indicating estimated effects of future weather conditions on the particular utility system is received. The information that indicates the estimated effects of future weather conditions on the particular utility system is provided for display.