The randomized MAC address doesn't help here. If two probe requests have different MAC addresses but the same SSID list, then the tracker can guess that they are the same device.
Each device sends beacons out at an interval. By sorting all the probes by these intervals (10Hz or what ever) each will likely be slightly different from each other. So my device sends probes out at 0s another will send it out at 0.5s. Also by co-relating these beacons by signal strength well the random MAC doesn't really matter.