In this work, we introduce a methodology that takes advantage of the inherent network diversity present in vehicular communications to improve the performance of safety applications. This methodology is based on a framework that simultaneously exploits the strengths of each individual network by using a set of decision rules. The implementation begins with a manual approach in which a typical, hierarchical decision tree characterizes the decision process of a single application when sending data to other users in the network. Analytical and simulation results validate the decision system approach when diversity is exploited as demonstrated by a boost in application performance, achieving an average latency under 100 ms and a 40% increase in throughput due to the increased packet delivery ratio. We then apply an ensemble learning technique, Random Forests (RF), to automatically reproduce the performance of the manually built tree system. Simulations under realistic traffic scenarios show the RF approach can replicate manually-built tree performance with up to 98% precision. A comparison with another state-of-the-art hybrid method also shows the RF scheme improves performance under a different application scenario without additional manual adjustments. With our methodology, we can add different application requirements and network characteristics to obtain a fully automated and adaptable decision system to optimize vehicular safety applications.