As power systems become more complex in order to accommodate distributed generation and increased demand, determining the stability status of a system after a severe contingency is becoming more difficult. To that end, artificial intelligence and machine learning techniques have been studied as a stability prediction tool. Topology changes and data availability however, impose certain limitations towards the generalization of those algorithms, impairing their ability to function in different system conditions. In this paper, we propose a novel ensemble machine-learning model that can maintain high performance in uneven sample class distribution, thus demonstrating resiliency and robustness against false dismissals.