One or more embodiments are directed multistage device clustering. A log including network traffic of multiple devices in a network is received. From the log, features of the devices are extracted and an aggregated feature matrix generated. A traffic behavior subset of the features in the aggregated feature matrix is selected, and a topic modeling algorithm applied thereto to obtain traffic behavior device groups. An application behavior subset of the features in the aggregated feature matrix is selected. On a per traffic behavior device group basis, the topic modeling algorithm is applied to the application behavior subset to obtain application behavior device subgroups. One or more devices are assigned to at least one of the plurality of application behavior device subgroups to obtain an assignment.