In one embodiment, a method includes accessing, by one or more computing devices, user sensor data from one or more wearable sensors on one or more players and optical sensor data from one or more cameras, where the user sensor data includes location data of the player and acceleration data, and where the optical sensor data includes several frames portraying the players and several scenes from an athletic event. The one or more computing devices analyzes, using a machine-learning model, the optical sensor data to identify the players and one or more actions during the athletic event and calculates one or more player metrics for the players based on the user sensor data and the identified actions. The one or more computing devices normalizes the player metrics for the players based on one or more weighted parameters and provides a report to one or more users.