Data from many buildings is used to machine train a cascade of classifiers. The cascade learns to classify in layers that relate to each other. One classifier identifies one characteristic and the other classifier uses the identified characteristic to identify another characteristic. For example, a cascade learns to classify buildings, efficiency, or cost associated with particular building automation systems (e.g., building A has increased heating cost), to classify the fault leading to the cost (e.g., hot water supply temperature causes increased cost), and to classify the source of the fault (e.g., valve position or valve) in a cascade. The resulting machine-learnt cascade is applied to data for any number of buildings.