At stressed operating conditions, critical contingencies can initiate loss of synchronism and trigger cascading events. Controlled islanding is the last line of defense to stabilize the whole system. This paper presents a decision tree assisted scheme to determine the timing of controlled islanding in real time by using phasor measurements. In addition, a slow coherency based approach is used to determine where to island. This scheme is tested on the operational model of the Entergy system and a severe N-2 outage case is used to demonstrate the phenomenon of cascading events due to protective relay actions. The results show that training one decision tree only for a specified critical contingency that can potentially cause cascading events can yields high prediction accuracy. Being aware of loss of synchronism in real time, operators can implement controlled islanding at carefully designed transmission interfaces and rapidly stabilize each island. Thus a significant amount of load is still served compared to uncontrolled system islanding.