DiffServ (Differentiated services) uses a 6-bit differentiated services code point (DSCP) in the 8-bit differentiated services field (DS field) in the IP header for packet classification purposes. If we already have that for QoS enforcement, why do we need machine learning for QoS prediction in Software Defined Networks? There are many papers that show machine learning mechanisms for QoS classification. Can't we just use the DSCP for that instead?
Something has to set the DSCP values. In a traditional network, that comes from either the end user / application, or intervening network hardware. Unless you have significant control of the device/application -- eg. VoIP phone -- the former can be a very dangerous road. The later is a very complicated, and tedious process. It is this area that machine learning can greatly demystify the voodoo of QoS.